Currency overlay 0470850272, 9780470850275, 9780470871638

Currency overlay has grown in parallel with the international diversification of institutional investment portfolios. Cu

270 30 2MB

English Pages 318 Year 2003

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

Currency overlay
 0470850272, 9780470850275, 9780470871638

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Currency Overlay Neil Record

Currency Overlay

Wiley Finance Series Currency Overlay Neil Record Active Investment Management Charles Jackson Option Theory Peter James The Simple Rules of Risk: Revisiting the Art of Risk Management Erik Banks Capital Asset Investment: Strategy, Tactics and Tools Anthony F. Herbst Brand Assets Tony Tollington Swaps and other Derivatives Richard Flavell Currency Strategy: A Practitioner’s Guide to Currency Trading, Hedging and Forecasting Callum Henderson The Investor’s Guide to Economic Fundamentals John Calverley Measuring Market Risk Kevin Dowd An Introduction to Market Risk Management Kevin Dowd Behavioural Finance James Montier Asset Management: Equities Demystified Shanta Acharya An Introduction to Capital Markets: Products, Strategies, Participants Andrew M. Chisholm Hedge Funds: Myths and Limits Francois-Serge Lhabitant The Manager’s Concise Guide to Risk Jihad S. Nader Securities Operations: A guide to trade and position management Michael Simmons Modeling, Measuring and Hedging Operational Risk Marcelo Cruz Monte Carlo Methods in Finance Peter J¨ackel Building and Using Dynamic Interest Rate Models Ken Kortanek and Vladimir Medvedev Structured Equity Derivatives: The Definitive Guide to Exotic Options and Structured Notes Harry Kat Advanced Modelling in Finance Using Excel and VBA Mary Jackson and Mike Staunton Operational Risk: Measurement and Modelling Jack King Advanced Credit Risk Analysis: Financial Approaches and Mathematical Models to Assess, Price and Manage Credit Risk Didier Cossin and Hugues Pirotte Interest Rate Modelling Jessica James and Nick Webber Volatility and Correlation in the Pricing of Equity, FX and Interest-Rate Options Riccardo Rebonato Risk Management and Analysis vol. 1: Measuring and Modelling Financial Risk Carol Alexander (ed.) Risk Management and Analysis vol. 2: New Markets and Products Carol Alexander (ed.) Interest-Rate Option Models: Understanding, Analysing and Using Models for Exotic Interest-Rate Options (second edition) Riccardo Rebonato

Currency Overlay Neil Record

C 2003 Copyright 

John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ, England Telephone

(+44) 1243 779777

Email (for orders and customer service enquiries): [email protected] Visit our Home Page on www.wileyeurope.com or www.wiley.com All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except under the terms of the Copyright, Designs and Patents Act 1988 or under the terms of a licence issued by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1T 4LP, UK, without the permission in writing of the Publisher. Requests to the Publisher should be addressed to the Permissions Department, John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ, England, or emailed to [email protected], or faxed to (+44) 1243 770620. This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold on the understanding that the Publisher is not engaged in rendering professional services. If professional advice or other expert assistance is required, the services of a competent professional should be sought. Other Wiley Editorial Offices John Wiley & Sons Inc., 111 River Street, Hoboken, NJ 07030, USA Jossey-Bass, 989 Market Street, San Francisco, CA 94103-1741, USA Wiley-VCH Verlag GmbH, Boschstr. 12, D-69469 Weinheim, Germany John Wiley & Sons Australia Ltd, 33 Park Road, Milton, Queensland 4064, Australia John Wiley & Sons (Asia) Pte Ltd, 2 Clementi Loop #02-01, Jin Xing Distripark, Singapore 129809 John Wiley & Sons Canada Ltd, 22 Worcester Road, Etobicoke, Ontario, Canada M9W 1L1 Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books.

Library of Congress Cataloging-in-Publication Data Record, Neil. Currency overlay / Neil Record. p. cm. Includes bibliographical references and index. ISBN 0-470-85027-2 (Cloth: alk. paper) 1. Currency overlay. I. Title. HG4529.5 .R43 2003 2003014721 332.4′ 5—dc22 British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN 0-470-85027-2 Typeset in 10/12pt Times by TechBooks, New Delhi, India Printed and bound in Great Britain by Antony Rowe Ltd, Chippenham, Wiltshire This book is printed on acid-free paper responsibly manufactured from sustainable forestry in which at least two trees are planted for each one used for paper production.

To Julie, Chris, Rob, Helen, Katy and Guy – a cure for insomnia!

Contents List of boxes

xv

Biography

xvii

Acknowledgements

xix

1 Introduction 1.1 Investment background 1.1.1 Investor instruments 1.1.2 Key investor categories 1.1.3 Defined benefit pensions 1.1.4 Defined contribution pensions 1.1.5 Investors in a currency overlay context 2 The Problem 2.1 Asset and liability valuations, volatility and solvency 2.1.1 Funded pension schemes 2.1.2 Asset valuations 2.1.3 Liability valuations 2.1.4 Liabilities’ discount rate 2.1.5 FRS17 2.1.6 IAS19 2.1.7 Summary on assets and liabilities 2.2 History of pension fund cross-border portfolio investing 2.2.1 US 2.2.2 UK 2.3 Currency volatility 2.4 Corporate parallels in cross-border investing 2.4.1 Foreign assets 2.4.2 Foreign debt 2.4.3 Economic impact of corporate currency exposure

1 2 2 3 5 6 6 9 9 9 9 11 11 13 13 13 14 14 14 15 20 20 20 21

viii

Contents

3 Currency Hedging 3.1 Instruments available 3.1.1 Foreign debt 3.1.2 Forward contracts 3.1.3 Currency swaps 3.1.4 Currency futures 3.1.5 Currency options 3.2 Option pricing 3.2.1 First principles 3.2.2 Option pricing theory 3.3 The Black–Scholes model 3.3.1 Market assumptions 3.3.2 The model 3.3.3 Understanding option pricing 3.3.4 The role of assumptions in option pricing 3.3.5 Practical implications of assumptions violations 3.4 Currency option pricing history 3.4.1 Lognormality 3.4.2 Monte Carlo models 3.4.3 Costs 3.4.4 Sensitivity 3.5 Interest rates and forward currency rates 3.6 Currency Surprise 3.6.1 What is currency surprise? 3.6.2 Currency surprise calculation 3.6.3 Why not spot returns? 3.6.4 Geometric linking and ‘adding across’

23 23 23 25 31 34 37 38 39 39 40 40 41 42 46 48 49 49 50 50 51 51 53 53 53 54 57

4 Foreign Exchange Market – History and Structure 4.1 A brief history of the foreign exchange market and how instruments developed 4.1.1 Bretton Woods 4.1.2 Central banks as buffer 4.1.3 Ad hoc foreign exchange market development 4.1.4 Free markets dominate 4.1.5 The euro 4.1.6 Instruments 4.2 Basic structure 4.2.1 Market size 4.2.2 Banks – the market-makers 4.2.3 Customers 4.2.4 Clearing mechanism 4.2.5 Turnover excluding ‘clearing’ 4.3 Customer types 4.3.1 Industrial and commercial companies (ICCs) 4.3.2 Oil and commodity dealers and merchants 4.3.3 Financial institutions (banks and insurance companies) 4.3.4 FX option writers

61 61 61 61 62 63 64 65 65 66 67 67 69 70 70 70 71 71 72

Contents

4.3.5 Investment pool traders in FX (hedge funds, proprietary traders) 4.3.6 Investment managers and currency overlay managers 4.3.7 Central banks 4.4 Physical and regulatory issues 4.4.1 Exchange controls 4.4.2 Taxation 4.4.3 Financial regulation

ix

72 73 74 75 75 79 80

5 Theory of Currency Hedging of International Portfolios 5.1 Lognormal random walk returns 5.1.1 Measurement 5.1.2 Returns 5.1.3 Volatility 5.1.4 Normally distributed period returns 5.1.5 A simple test 5.1.6 Relevance to currency hedging 5.2 The ‘free lunch’ 5.2.1 Which way up? 5.2.2 Adding ‘moving parts’ 5.2.3 Currency exposure is different 5.2.4 Correlation of asset classes 5.3 Hedging and the efficient frontier 5.3.1 Constructing an optimiser including currency 5.3.2 Optimiser methodology 5.4 Implications of transactions costs 5.4.1 Expected portfolio added value from passive hedging

83 83 83 84 86 88 89 89 90 91 91 93 94 97 97 98 101 102

6 Passive Currency Overlay 6.1 Mechanics 6.1.1 Original maturity of forward contracts 6.1.2 Frequency of cash flows 6.1.3 Currencies to be hedged 6.1.4 Benchmark or actual asset weights to be hedged? 6.1.5 Denominator of ‘contribution from hedging’ 6.1.6 Frequency of asset valuation 6.2 Rebalancing 6.2.1 Frequency of rebalancing 6.2.2 Rebalancing buffer (Y/N? size) 6.2.3 Buffer – ‘percentage of what?’ 6.2.4 Delay in rebalancing 6.2.5 Valuation rates 6.3 Cash flow 6.4 Costs 6.4.1 Direct costs 6.4.2 Indirect costs 6.4.3 Summary on costs 6.5 Postscript on costs – conflict of interest

105 105 105 106 106 108 110 110 111 112 114 115 115 116 116 117 118 119 121 122

x

Contents

7 Currency Overlay Benchmarks 7.1 What is a currency benchmark? 7.1.1 Misleading currency attribution 7.1.2 Benchmark as portfolio 7.1.3 Benchmark mechanics 7.2 Investability 7.2.1 Forward currency prices 7.2.2 WM/Reuters rates 7.2.3 Contract rolling 7.2.4 Scale of contracts 7.2.5 Rebalancing 7.2.6 Geometric linking 7.3 Design 7.3.1 Asset plus currency overlay methodology 7.3.2 Special case – monthly benchmark calculation 7.3.3 Valuation of unmatured contracts 7.3.4 Benchmark hedge ratio 7.3.5 Embedded currency plus currency overlay methodology 7.3.6 Currency overlay only methodology 7.3.7 Other methodologies 7.3.8 A currency benchmark with or without asset returns? 7.3.9 Pricing/costs 7.3.10 Prices, not interest rates 7.3.11 Rebalancing 7.3.12 Original contract maturity 7.3.13 Constant maturity benchmarks 7.3.14 Discounting 7.3.15 Benchmark hedge ratio – strategic considerations 7.3.16 Currency coverage and denominator calculation 7.3.17 Underlay 7.3.18 Benchmark performance 7.3.19 Benchmark cash flows 7.4 Current practice 7.5 Worked examples 7.5.1 Asset plus currency overlay methodology 7.5.2 Embedded currency plus currency overlay methodology 7.5.3 Currency overlay only 7.6 Tracking error 7.6.1 Passive hedging 7.6.2 Summary on tracking error

125 125 126 126 126 127 127 128 129 129 129 129 130 131 132 133 133 133 134 134 135 136 136 137 138 138 139 139 140 140 140 141 144 145 145 147 148 148 149 155

8 Overlaying Different Asset Classes 8.1 Equities 8.1.1 Correlation – the historical evidence 8.1.2 Correlation evidence 8.1.3 Individual currency: equity correlations 8.1.4 Stability of correlations

157 157 157 158 159 160

Contents

8.2 8.3

8.4 8.5

8.1.5 Summary on correlation 8.1.6 Embedded currency 8.1.7 Firm level analysis 8.1.8 Country index equity returns 8.1.9 International equity correlations 8.1.10 Volatility reduction – the historical evidence 8.1.11 Effect of hedging on portfolio risk 8.1.12 Base-currency-specific graphs Hedge ratios 8.2.1 Current debate Bonds 8.3.1 Correlation 8.3.2 Stability of correlations 8.3.3 Volatility reduction from hedging bonds – the historical evidence 8.3.4 International diversification Property Other classes

xi

162 162 162 167 169 170 172 174 177 177 182 182 185 187 187 191 191

9 Is the Currency Market Efficient? 9.1 Types of inefficiency 9.2 Making the case for currency market inefficiency 9.2.1 Cyclical behaviour 9.2.2 Lack of statistical arbitrage 9.3 Empirical evidence for medium-term trends 9.4 Forward rate bias – another inefficiency 9.4.1 What is the evidence for the FRB? 9.4.2 Risk premium 9.4.3 Monetary policy and inflation 9.4.4 Nominal rate illusion 9.4.5 Other inefficiencies 9.5 A successful universe? 9.5.1 An example of different perspectives in the FX market 9.6 Summary on evidence for inefficiency 9.6.1 Weak form efficiency 9.6.2 Semi-strong form efficiency 9.6.3 Strong form efficiency 9.6.4 Transactional efficiency

193 193 194 194 194 195 198 199 199 201 202 202 202 203 203 203 204 204 204

10 Active Currency Overlay – Management Styles 10.1 The problem 10.2 Modelling and forecasting 10.2.1 Modelling – Occam’s razor 10.2.2 Can models work? 10.2.3 What about active management without models? 10.2.4 Dealing and practical execution 10.2.5 Timeliness of inputs

205 205 206 206 207 209 210 210

xii

Contents

10.2.6 Judgement and modelling 10.2.7 Deal execution 10.3 Active management styles 10.3.1 Fundamental 10.3.2 Technical 10.3.3 Option-based 10.3.4 Dynamic

211 212 212 213 219 229 234

11 Active Currency Overlay – Evidence of Performance 11.1 Surveys 11.1.1 Currency overlay performance surveys 11.1.2 Performance summary 11.2 Who loses?

239 239 239 240 240

12 Implementing Currency Overlay 12.1 Summary check-list 12.1.1 What mandate type? 12.1.2 For risk-reducing overlay – benchmark hedge ratio 12.1.3 Investment guidelines – active 12.1.4 Investment guidelines – passive 12.1.5 Investment guidelines – alpha 12.1.6 Bank FX lines 12.1.7 Bank contract confirmation 12.1.8 Investment management agreement 12.1.9 Reporting requirements 12.1.10 Periodic cash and contract reconciliation 12.1.11 Bank contract settlement procedures 12.1.12 Benchmark calculation 12.1.13 Performance measurement 12.1.14 Summary check-list 12.2 Practical questions and answers

243 243 243 244 245 246 246 247 248 248 248 249 249 249 250 250 251

13 Looking Ahead 13.1 Development of active management styles 13.1.1 Top-down/bottom-up 13.1.2 Growth/value 13.1.3 Contrarian/momentum 13.1.4 Ethical 13.1.5 Hedge fund styles 13.1.6 Summary on styles 13.2 Natural selection of overlay managers 13.2.1 Conflict of interest 13.3 Extending the range of hedging instruments 13.4 Will inefficiencies grow or shrink? 13.4.1 Can outperformance by currency overlay managers continue?

253 253 253 254 254 254 255 255 256 257 258 258 259

Contents

xiii

References/Useful Reading

261

Appendices Appendix 1 – Boundary conditions for forward arbitrage Appendix 2 – Lognormal returns Appendix 3 – AIMRR report Appendix 4 – Sample investment guidelines

263 263 267 277 285

Index

289

List of Boxes 1.1 What is currency overlay? 1.2 Life expectancy 1.3 Defined benefit pensions

2 4 5

2.1 2.2 2.3 2.4 2.5 2.6

Discounting Currency surprise Swift codes Volatility The euro Swaps

12 16 18 20 21 22

3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8

Forward currency contracts FX swaps Bid/offer spreads Futures Currency options Normal and lognormal Forward currency rates Local currency returns

25 26 29 34 38 41 52 56

4.1 4.2 4.3 4.4 4.5 4.6 4.7

Capital flows Currency unions Netting Market-makers CLS Bank Exchange controls Exchange controls – UK case study

62 64 67 68 69 77 79

5.1 5.2 5.3 5.4

Convexity Central Limit Theorem Siegel’s paradox Fat tails

84 85 87 88

xvi

List of Boxes

5.5 Mean-variance optimisation 5.6 Modern portfolio theory

97 99

6.1 Tracking error 6.2 Tradeable currency 6.3 Proxy hedge

107 108 109

7.1 Benchmarks 7.2 MSCI EAFE

125 144

8.1 Embedded currency exposure 8.2 Duration

163 190

9.1 Trends 9.2 Forward rate bias 9.3 Currency risk premium

197 198 200

10.1 10.2 10.3 10.4 10.5 10.6 10.7 10.8 10.9

Cross hedging Alpha ‘Star’ investors Active currency styles Capital account Purchasing power parity Mean reversion Information ratio Simulations

205 208 209 212 214 216 224 225 227

Biography Neil Record was educated at Balliol College, Oxford, where he studied Philosophy & Psychology; Essex University and University College, London, from where he gained an MSc in Economics with Distinction. His first job, in 1977, was as an Economist in the Economic Intelligence Department of the Bank of England. While there he had his first taste of largescale economic modelling, and his first exposure to an exchange rate forecasting model. It was notable then that the exchange-rate forecasting model was one of the weaker elements of the Bank’s model of the UK economy. In 1979, Neil moved to work as a commodity price forecaster at the chocolate maker Mars in Slough, UK. He quickly moved into line responsibility for buying commodities, and then to buying forward currency for the company’s commodity and import needs. While at Mars he developed an innovative process for controlling currency risk and exploiting currency market inefficiency. In 1983, aged 29, Neil left Mars to establish his own specialist currency management business, Record Treasury Management (renamed Record Currency Management in 2001). Record Currency Management was a pioneer of currency overlay for pension and investment funds, and in April 1985 it was awarded the worldwide first-ever institutional currency overlay mandate – from the UK Water Authorities’ Superannuation Fund. Today, 20 years after founding Record Currency Management, Neil remains Chairman and CEO. He divides his time between client liaison, currency overlay design work, and speaking and writing about currency overlay. His other interests include a non-Executive Directorship of RDF Media – a leading UK independent TV production company – opera, baroque music, tennis, skiing, travel, art, architecture and watching his childrens’ sporting activities. He is divorced and lives with his partner, Julie, and his three children and her two children.

Acknowledgements I am indebted to my colleagues at Record Currency Management, particularly Peter Wakefield, Leslie Hill, Bob Noyen, Mike Timmins, Ian Harrison, Joel Sleigh, Dave Murphy, Simon Williamson, James Dyas, Sebastian Jans and Dmitri Tikhonov for their technical input, advice and debate. I am also grateful to Chris Jackson, Jane Deane and Julia Edbrooke for their administrative support. I am also indebted to Brian St.J. Hall of Hewitt Bacon & Woodrow for advice and comments on earlier drafts. Finally, I must thank Julie for putting a glass of wine in my hand after particularly heavy days!

1 Introduction Currency overlay is a new branch of investment management. Not surprisingly, the literature is equally new; in particular, there does not exist at the time of writing (2003) any comparable comprehensive book on currency overlay. Because currency overlay is new, I have chosen to write this book in such a way that it fulfils two functions: the first as a primer; the second as a reference book. Writing for a heterogeneous audience is difficult at the best of times; writing for a heterogeneous audience and to fulfil two functions is doubly challenging. I have chosen to enlist a number of techniques to help in this effort.

r Argument from first principles

r

r

Where possible, I try to limit the reference to external theories or academic literature. In theory, I would like the able and mathematically literate reader to be able to grasp the arguments in this book without having prior knowledge of the currency world or any of the literature that surrounds it. Use of mathematics I will use only sufficient mathematics to explain my point or make my case. I have avoided advanced maths where at all possible. Generally, the only references to ‘given’ mathematical theories are in statistical formulae, where there is simply not enough space to derive them from first principles. Boxes I will use a technique used by Norman Davies in Europe – a History,1 in which he creates ‘boxes’ with self-contained ‘stories’. While a full description of currency overlay is clearly nowhere near as multi-stranded as the history of Europe, nevertheless I believe the reader will find it helpful to have text which is not fully ‘linear’. I also use boxes to explain or describe vocabulary which may be unfamiliar.

This book is aimed at a wide audience. I anticipate that it will include readers from currency overlay managers, from the wider investment management industry, from investment consultants and actuaries, from the more investment-literate pension fund managers, from the investment management desks of insurance companies, from foreign exchange dealing banks and from custodian banks. It will inevitably also appear on the desks of students in finance and related disciplines, although it is not written as a textbook. What distinguishes this book from a similar one written as a textbook? Firstly, it is not designed to fit in with any class, course, degree or qualification. Secondly, it is not designed to be fully rigorous. This is deliberate – full rigour would significantly expand the size of the book without adding to the core content. Thirdly, a textbook would not typically aim at a heterogeneous audience. A readership of, say, graduates following an AIMR2 qualification or a 1

Davies, N., Europe – a History, Oxford University Press, December 1996; ISBN 0-195-20912-5. Association for Investment Management & Research – the highly respected American asset management trade association and the awarder of the CFA qualification. 2

2

Currency Overlay

postgraduate degree course would expect a great deal less introductory finance and economics, and a great deal more mathematical rigour. Currency overlay is not, to my knowledge, part of the syllabus of any investment management or other finance course at the moment. When, and if, it becomes part of standard graduate or professional courses, a textbook will undoubtedly emerge.

1.1 INVESTMENT BACKGROUND Readers completely unfamiliar with investment may like a brief primer as to the key investment instruments and the key investor categories. I do this because I will be referring to these instruments throughout the book. The reader familiar with basic investment principles might like to skip this short section, and turn to p. 14. 1.1.1 Investor instruments There are two main investment classes: equities and bonds, and both of these are also instruments. There is a larger range of smaller asset classes (cash, property, private equity, hedge funds, commodities), although some of these (e.g. hedge funds) are not instruments. Box 1.1 What is currency overlay? An investor – say a US investor – decides to invest in the UK stock market. The decision to invest in the UK is not determined by the investor’s view on the prospects for the UK pound, but by the good prospects for the UK stock market and a desire for diversification. However, to buy the UK stocks, the US investor has to buy pounds with dollars on the foreign exchange market. He then buys the UK stocks with the pounds. Each month, the investor will get a valuation of this investment, and two markets (not one) will affect this valuation – (i) the general level of UK stocks in pounds and (ii) the level of the pound against the dollar. The investor may decide that the uncertainty brought about by the variation in the dollar/pound exchange rate is undesirable, and choose to get rid of (or ‘hedge’) it. He hires a currency overlay manager, whose remit is to maintain a portfolio of forward currency contracts to offset any movements in the dollar/pound exchange rate. Broadly speaking, these contracts commit the investor to sell fixed amounts of pounds against the dollar at fixed dates in the future. This ‘overlay’ does not require any investment on the part of the investor; only a credit line with foreign exchange banks to enable the overlay manager to have the ability to deal in forward contracts with the banks on behalf of the investor. The ‘returns’ of currency overlay come in two ways: forward currency contract valuations and cash settlements on maturing forward contracts. These two combine to become the contribution to return of currency overlay. The overlay manager who maintains the currency hedge over a constant proportion of the portfolio is conducting ‘passive overlay’; the manager who varies it according to his view on the prospects for the pound is running ‘active overlay’. The expression ‘currency overlay’ has been extended recently to cover active mandates which are purely return-seeking, rather than risk-reducing. These mandates are not restricted purely to forward contracts to sell currencies already owned in the portfolio against the base currency – they can utilise any currency pair in either direction.

Introduction

3

Equities (or shares) are a share in a limited liability company, and entitle the holder to the residual value on winding up of a company, and to regular payments in the form of dividends of surplus profits after other claims on the business have been discharged. Quoted equities (those that trade on recognised exchanges) are generally sufficiently liquid to maintain a ‘secondhand’ price in the market. This market price is the key for much of the later discussion in this book about international equities and currency risk. For the time being it is worth noting that the currency in which equities are quoted (which is generally, but not exclusively, the same as the currency in which the company reports its accounts) is important for the initial categorisation of equities’ currency risk. Bonds are on the face of it much simpler instruments than equities. They are a loan from the investor to a company or other economic entity. Bond-issuing entities include companies, asset-holding special purpose vehicles, governments, international organisations and mixed public/private sector groups. The bond (and I find it easiest even in this electronic world to think of it as a piece of paper) represents a promise of a series of payments by the issuer to the holder. These payments are a fixed amount of cash (the coupon), say $6 per year, until a particular date, and then a payment of principal, which is almost always by convention $100 (or £100 or Eur100 depending on the currency of the bond). Bond coupon payments are usually made either annually (mainly governments) or semi-annually. For annual payments, the last payment date (when the principal is repaid) is also usually the payment date of the coupon. There are a huge range of variations on this simple theme – most importantly that there are bonds where both the coupon and the principal are linked to an index of consumer or retail prices (‘index-linked’ bonds), and where the amount of the coupon varies according to prevailing short-term rates, rather than being fixed (‘floating rate’ bonds). Bonds can be and are issued in a variety of currencies, although they are for practical purposes traded (in the second-hand market) in the same currency as their issue currency. 1.1.2 Key investor categories Who are the main investors in equities and bonds? The ultimate answer is ‘individuals’, but most of this ownership is diffused through one or more intermediary layers, so that the majority of the influential decision-making is made by the managers of large pools of assets. These are commonly called ‘institutional investors’, although this is something of a misnomer, since they are mainly professional commercially-orientated managers, rather than representatives of ‘institutions’. Investment is about deferring consumption. There are many different reasons for deferring consumption, but by far the most important now is the deferral of income from the period while an individual is working to their retirement. This used not to be the case, at least in public securities’ markets. In the heyday of nineteenth century capitalism, the majority of investors were rich men intent on getting richer – not thinking about their retirement, and insurance and provident companies investing small contributions from policyholders to provide for life’s disasters – death, injury, fire – but not generally retirement. Few women in this period were economically active (in the sense of having paid jobs), and men were expected to work until 60 or 65. The fundamental changes in the economic and social structure of western societies in the late twentieth century have changed that – families no longer expect to look after their elderly relatives, life expectancy in the developed countries has been extended markedly and average retirement age has, if anything, gone down. See Box 1.2. It is not untypical for a

4

Currency Overlay

worker to retire in their mid to late fifties and live to their early eighties. This requires 25 years of income – income that governments are either unwilling, or now increasingly unable, to provide. Providing this income is the principal concern of the investment management industry. Box 1.2 Life expectancy Life expectancy in developed countries has been increasing almost continuously for two centuries. However, the pattern of mortality has also changed, and this has a significant impact on the need and scale of retirement provision. Taking the UK as an example, the striking thing about the change in mortality patterns is the collapse in early death rates. In the nineteenth and early twentieth centuries, people died at every age group – mainly through infectious diseases and what would be regarded now as avoidable causes – accidents, childbirth, etc. In the late twentieth and early twenty-first centuries, death is almost exclusively the preserve of the elderly. This means that the population is becoming increasingly elderly, and the ratio of working years to retirement years has collapsed. In 1841, only 30% of males born reached 65; in 2000, 83% reached 65. Life expectancy for those lucky enough to reach 65 has changed less markedly. In 1841, the life expectancy of a 65-year-old man was 76 years; in 2000, 81 years. This effect can be seen graphically in male life expectancy at ages 0, 45 and 65: Male life expectancy 90.0

Life expectancy

80.0 70.0 60.0 at 65 50.0

at 45 at birth

40.0 30.0 1841*

1870*

1900*

1930*

1960**

1990***

2020p

Year

Source : UK Government Actuary's Department.

The female experience is very similar: just a few years longer at every age. So both the continuous rise in the solid line, and the recent rise in the dotted and dashed lines, are making the provision of retirement income so much more difficult and material. These changing social and economic trends mean that the majority of investment is by pension funds. While there is a large sector which is explicitly labelled ‘pension funds’, there are also mutual funds and insurance company investments which, even if not labelled so, are also largely for the purpose of securing individual income in retirement. Pensions are divisible into two key types – defined benefit and defined contribution.

Introduction

5

1.1.3 Defined benefit pensions Defined benefit (DB) pensions are just that. They are a promise by an employer, or a government, to pay a defined amount of annual pension between a retirement date and death. The Box 1.3 Defined benefit pensions The key characteristic of defined benefit (DB) pensions is that the pension they deliver is not affected by any factors except the number of years of pensionable employment and the employees’ final (or, more rarely, average) salary. DB pensions are typically expressed in the form ‘1/60 of final salary for each year of pensionable employment’. In the UK, but not the US, they are now required to be index-linked (0–5% p.a. limits) after retirement. They are typically offered to men and women on the same terms, although the expected cost to the employer is higher for women than for men because of the higher life expectancy for women. The cost of DB pensions is highly sensitive to real investment returns, salary increases and longevity. If the cost of providing a DB pension were outsourced to a third-party provider like an insurance company, then it would be possible to calculate from financial economics the annual cost to the employer that the insurance company would have to charge. I assume here that the insurance company is prepared to take mortality (annuity) risk (which is diversified), but not investment risk (which is systematic), and that the contract is priced at cost. The graph below shows the annual pension cost compared to employee age on the following assumptions: UK pension; investment only in I/L gilts; 2.5% real returns for I/L; I/L pension (no cap or floor); UK average mortality; annual salary increase 3% p.a. higher than inflation; continuous employment with one employer for 40 years; retirement at 60; 1/60th accrual. Later years are so much more expensive to the employer because of the uprating of accrued rights with salary increases. And just to be clear, this graph means that the annual cost to the employer of a male aged 59 is 60% of his salary! Risk-free cost of DB pension by age (% of annual salary) 70%

60%

Males Females

50%

40%

30%

20%

10%

0% 20

25

30

35

40

45

50

55

60

On these assumptions, the average cost per year to the employer is 26% p.a. for males, and 30% p.a. for females.

6

Currency Overlay

defined amount may go up (in the UK it routinely increases with inflation for example), but it will never go down. The level of pension is not dependent on the performance of any underlying assets guaranteeing the pension – only on the rules of the pension. This is the distinguishing feature from defined contribution pension schemes. 1.1.4 Defined contribution pensions Defined contribution (DC) pension schemes are not really pension schemes. They are basically mutual funds or unit trusts with a beneficial tax-wrapper. The ‘tax-wrapper’ is usually that governments will allow income tax relief on investment into a DC scheme, and allow the income and capital gains to roll up inside the fund tax-free.3 By way of compensation, governments usually tax all of the subsequent capital value as income when it is received by the pensioner, which will usually be by way of annuity receipts. DC pension schemes allow individuals to build their own ‘pot’ of assets, often with contributions also coming from their employer. Most jurisdictions require retiring DC members to buy an annuity with much or all of this ‘pot’. An annuity is a promise from a provider (usually an insurance company) to pay a fixed annual amount from the date of purchase until the annuitant’s death. Payments from an annuity are, in effect, a pension. Annuities are priced by the market, which is run by insurance companies. They pool the mortality risk they are taking, and then replicate the payments required by holding bonds. If bonds are expensive (i.e. bond yields are low), then annuities will be expensive (i.e. a fixed outlay will buy less annual pension). The DC investor therefore runs two very significant financial risks: (1) the risk of poor performance of his investments up to the point at which he retires, and (2) the risk that annuities will be expensive, thereby reducing his pension. 1.1.5 Investors in a currency overlay context The remainder of this book will deal with the issue of currency exposure and institutional investors. The currency exposure at the centre of this question is that embedded in the ownership of securities outside an investor’s home country, and denominated in a foreign currency. As a practical matter, most of the currency overlay mandates awarded in the period 1985–2002 have been to manage the currency exposure of international equity portfolios. In addition, most of the investors awarding such contracts have been DB pension funds. This has come about because the majority of cross-border portfolio investment is experienced by DB schemes’ equity portfolios. Table 1.1 shows rough estimates of the scale of global asset classes. It is very difficult to make sensible estimates of total cross-border asset holdings, hence my choice to show estimates for DB cross-border holdings alone (which are better documented). Suffice it to say at this stage that international equities are the largest creator of identifiable currency risk, and DB pension funds are the largest of the investor types experiencing currency exposure.

3 In the UK, the notorious 1997 withdrawal of ACT relief for pension fund holders of UK equities was a partial withdrawal of this tax-free status. The position with regard to bonds is now anomalous – bond interest cost is fully tax-relievable for UK corporates, and not taxable in the hands of pension funds.

Introduction Table 1.1 Estimates of global asset ownership ($ bn) Total world market cap

Of which owned by DB funds

Of which DB cross-border

DB% total

Crossborder% DB

Equities Bonds

23 800 18 000

5200 3466

1126 125

22% 19%

22% 4%

Total

41 800

8666

1251

Sources: Lehman Bros; MSCI; Record Currency Management, December 2002.

7

2 The Problem 2.1 ASSET AND LIABILITY VALUATIONS, VOLATILITY AND SOLVENCY 2.1.1 Funded pension schemes Pension funds as distinct financial entities are a post-war phenomenon. They emerged in scale in the 1950s in the US and UK, and they largely invested in domestic, mainly government, bonds. Domestic bonds were regarded as a safe investment, and trustees, in many jurisdictions hedged around with legal limitations, had neither the expertise nor the incentive to invest in any other instruments. The idea of establishing trustees in a statutory framework to oversee the investment of a separate fund was gradually enshrined in legislation in the US,1 UK, Australia, Canada, Netherlands and Switzerland, as well as several smaller economies. Notably, the major economies of continental Europe, Germany, France and Italy, did not adopt this model, and to a great extent have not done so to this day. Japan is a special case, with substantial funded schemes, but not in the standard employer-sponsored model. Pension funds were only established because the sponsoring employer had made a pensions promise to its staff. Generally speaking, these promises consisted of the employee ‘earning’ a proportion of his final salary for each year of employment. A typical pension scheme might offer 1/60 or 1/80 of the final salary as pension for each year’s service. In the US this remains a promise which is not linked to inflation after retirement; in the UK defined benefit pensions are now required to be given limited price indexation – in effect linked to retail prices with a 5% p.a. cap and 0% p.a. floor. Governments and regulatory authorities were rightly wary of allowing companies to take on long-tail financial obligations to employees (who are also voters!) without putting in place some mechanism to ensure that these promises were fulfilled. Hence the pension fund, legally distinct from the company, and able to survive the insolvency of the company without danger of attack by the liquidator. Clearly, the desirable aim for the security of pensioners and future pensioners is to ensure that in the event of the insolvency of the sponsoring company, the assets in the pension fund are sufficient to pay all the outstanding promises in full. This opens the door to an area of significant controversy. I feel that in order to fully explore the concept of investment risk in a pension fund, I need to lay out some elements of that controversy. 2.1.2 Asset valuations In the modern world of electronic communication, real-time market price screens and large and liquid securities’ markets, it might seem that the question of establishing how much the assets of the pension fund are worth was a relatively trivial one. Why not look up the price of all the 1 Michael Clowes, editor for many years of the premier pension trade journal (Pensions & Investments) in the US, has written a detailed post-war history of US pension funds. Clowes, M. J., The Money Flood, John Wiley & Sons; ISBN 0-471-38483-6.

10

Currency Overlay

constituent stocks, and sum the consequent value of each holding over the whole portfolio? While there might remain some question over the pricing of unquoted securities and of property, nevertheless the realisable value would be established with a narrow margin. This procedure has now been adopted in many, but not all, of the major funded pension jurisdictions. Even in those where market valuation is now the standard, this change has in many cases happened within the last 10 years. What can possibly have been the alternative basis? The answer to this is dependent on the jurisdiction, and the practices of the profession responsible for pension fund valuation – actuaries. In the UK and the US, until fairly recently, many pension fund asset valuations were made on the basis of ‘discounted cash flow’. Discounted cash flow valuation ignores market values of otherwise perfectly marketable securities, and instead values them on the basis of a valuation model. As this applies to equities, the model runs something like this:

r Step 1 r Step 2 r Step 3 r Step 4

Establish the current dividend yield of the security or index Project the dividend out into the future using a dividend growth assumption Calculate the stream of dividend cash flows arising from these assumptions Discount the stream of cash flows to a present value (PV) at an arbitrary discount

rate

r Step 5

Call this PV the valuation of the assets

There are numerous objections to this valuation methodology, not least that the valuation so devised will not be the asset sale value. Other objections include the arbitrary nature of the dividend growth assumptions and the discount rate, neither of which can be readily determined by reference to a market. Indeed, it could be argued that the only possible reference to the market would be to derive an implied dividend growth rate and equity dividend discount rate from the market’s valuation of the security – which then makes the whole exercise plainly circular. Discounted cash flow valuations have all but been replaced by market valuations in pension fund valuations (see Table 2.1), so the discussion may seem a little academic. However, the culture that created actuarial valuations for assets lives on in the valuation of liabilities. Table 2.1 shows that the market value valuation methodology, while the largest category of asset valuation, is by no means the only methodology. ‘Smoothed value’ (taking only a proportion Table 2.1 Asset valuation method relative frequency in North America US Small plans Number of responses Asset valuation group Fair market value Discounted cash flow Book value Smoothed value Other (including combination of methods)

Canada Large plans

Small plans

Large plans

5799

3168

274

311

65.3% 0.0% 27.8% 6.9% 0.1%

48.6% 0.1% 13.9% 36.4% 1.0%

90.5% 0.0% 1.1% 8.0% 0.4%

47.3% 0.3% 4.5% 42.1% 5.8%

Source: Survey of Asset Valuation Methods for Defined Benefit Pension Plans, Society of Actuaries, 1999.

The Problem

11

of the asset value changes each year) and ‘book value’ (essentially cost) are also remarkably widespread. It is worth noting that ‘During the period from 1988 through 1996, plan assets were “marked to market” sparingly in the U.S. (a low of 0.3% of all plans in 1989 to a high of 2.6% of all plans in 1996) and very rarely in Canada’ (SOA 1999 Survey). So the change to market valuation has come very recently and very rapidly. The reader will notice ‘book value’ as a valuation alternative, which is essentially asset purchase cost (or possibly market valuation if less). Since most assets will have been long-held, book value will not acknowledge any volatility in valuation brought about either by equity market fluctuations, or by currency fluctuations. 2.1.3 Liability valuations Liabilities do not on the face of it fall as neatly as assets into market-based valuations. DB pension funds’ liabilities to future cash outflows are not certain. They are subject to a number of uncertainties, including:

r Mortality (currently on an unfavourable trend) r Pensions-in-payment inflation (depending on the pensions promise) r Salary inflation prior to retirement r Early leavers (a favourable risk to the fund) r Early retirement (variable effect) These factors combine to make the calculation of future cash flows an educated guess, or a range of values with probabilities assigned, rather than an exact value. However, all of these uncertainties can be estimated with more or less precision, and the estimates are unlikely to change rapidly (mortality), or can be hedged (inflation). One key uncertain variable is left to bring these future cash flows back to today’s values – a discount rate. 2.1.4 Liabilities’ discount rate The value, and the underlying calculative principles, of the discount rate are by no means a settled question, and indeed at the time of writing have been raised right to the top of the political agenda in the UK, and to a lesser extent in the US and elsewhere. What is so contentious? There has been a considerable history around the subject of liabilities’ valuation. The position in most of the developed economies with funded pension sectors is that liabilities are valued in two or three different ways depending on context. In the US, funds are required by ERISA2 to undergo a regular solvency examination to determine whether they meet minimum solvency requirements. This was the context of the asset valuation results shown in Table 2.1. The valuation of liabilities is similarly varied, with no pension funds reporting their liabilities discounted at the risk-free discount rate. In addition, companies are required to report their net liabilities and pension expense under the US accounting standard FAS87. However, this allows for smoothing of asset valuations, as well as a choice of three key variables for the liability valuation – salary inflation, ‘expected rate of return’ and the discount rate. In FAS87 parlance, the discount rate is commonly applied to pensions in payment, and to a lesser extent future pensions at the shorter end of the maturity spectrum. The ‘expected rate of return’ is the discount rate applied to longer maturity liabilities. The nomenclature is revealing: 2

The US Employee Retirement Income Security Act 1974.

12

Currency Overlay

liabilities do not have an expected rate of return; assets do. Why should the returns of a fund’s assets affect the calculation of the PV of the liabilities? Box 2.1

Discounting

A pension fund has obligations to pay its pensioners out into the distant future. How does the fund value this liability in today’s money? This is important if only because it gives a guide as to how much it needs in assets valued today to cover the liabilities. Assume for the moment that a pension fund owes a level £10m per year from today to 2050, after which it has no further obligation. (Pension liabilities do not have this level payment characteristic – it is for illustration only.) We can calculate a PV of this liability using a risk-free rate (say 4% p.a.) – the rate of return from investing in government bonds of the appropriate maturity. We can also value it at higher discount rates, representing a ‘hope’ that the assets in the fund (on the other side of the balance sheet) will yield a return yield than or equal to the higher assumed rate. We can show the huge impact of changing the discount rate in the following table: Discount rate

PV (£m)

% risk-free PV

4% (risk-free) 5% 6% 7% 8% 9% 10%

222 191 167 147 132 119 109

100% 86% 75% 66% 59% 54% 49%

The answer, I believe, lies in the realpolitik of the actuaries’ profession. Profit-maximising companies wish to minimise their payments to their pension funds, commensurate with the funds remaining solvent. As they increased their holdings of equities in the 1950s and 1960s, the superior returns of equities meant that each year the discount rate used for the valuation of their funds’ liabilities looked over-conservative. If their assets went up, on average, faster than their liabilities, then was there not a case for adjusting the discount rate of the liabilities to more closely match the returns achieved in the investment portfolio? The pressure from companies on their actuaries, and then from actuaries to government regulators (via their position on working groups, etc.) began to enshrine in law the idea that a risk-free rate for the liabilities’ discount rate was a luxury they neither needed nor could afford. In its place was put a series of discount rates, each applying to different types of liability (pensions in payment, active staff members, deferred pensioners (ex staff), roughly reflecting the maturity of the liability and its ‘certainty’. ‘Uncertain’ liabilities, particularly those of young active staff members, could be discounted at ‘assumed equity’ returns (in 2000, these were about 9% p.a. in the UK, versus a risk-free rate of about 5.5%). What has escaped most governments (despite a regular trickle of underfunded pension schemes in bankrupt companies) is that enshrining ‘long-term’ rates of return in the liability discount rates ensures that there is a strong likelihood that the primary purpose of pension funds – to secure promise payments to pensioners – will fail. The turn of the twentieth to the twenty-first century has seen interest rates (and inflation) across major industrialised countries fall to 40-year lows. This has not only created well-publicised problems for insurers like

The Problem

13

Equitable Life in the UK (who promised (high) Guaranteed Annuity Rates to some of its savers), but has also made the equity return assumption for the liability discount rate increasingly untenable. 2.1.5 FRS17 A new accounting standard, introduced in the UK in 2001 and originally coming fully into force in 2003 (but now delayed until 2005 to mesh with the move to international accounting standards), requires UK companies to put the assets of their pension fund less pension liabilities on the parent balance sheet (in the interim, in the notes). Not only is this the first time that UK companies have had to do this, but they are required to value assets at realisable market value on the balance sheet date, and liabilities at a discount rate equal to the AA corporate bond rate prevailing on the balance sheet date. There is no ‘smoothing’ and no ‘assumed rate of return’. This standard has almost always shown that pension funds reporting in 2001 and 2002 are in deficit, and sometimes by very large absolute and relative amounts. Bizarrely, the new standard, which goes some way in allowing the owners of companies to know what their net pension obligations are, is under attack from an unholy alliance of unions, business leaders and politicians. The unions want the standard postponed or abolished because they fear the bad news it brings will prompt the company to abandon final salary pension schemes; businessmen want it postponed or abolished because it introduces an ‘uncontrollable’ element of volatility onto their balance sheets, and they can’t afford to eliminate the volatility by investing in bonds; and politicians of all hues want it postponed or abolished because they fear it will lead to the closure of final salary pension schemes, which will not only be unpopular amongst a significant proportion of the electorate, but will also put more pressure on the state and the taxpayers to fund retirement. Those in favour are a small group of informed accountants and actuaries, who are attempting to provide financial transparency and objectivity to the users of company accounts – the equity holders. Ironically, pension funds form a significant proportion of equity holders, and they have remained mute in this role. This group will most likely win the battle – the Equitable Life debacle in the UK, and Enron in the US, very different in both cause and result, will ensure that the call for transparency is going to be politically irresistible. 2.1.6 IAS19 The UK and other EU members are committed to adopting international accounting standards by 2005, and the US will wish to move to closer harmony with these if at all possible. US and international standards currently allow smoothing and non-market liability discount rates – the outcome of the next few years’ debate will determine whether the new UK principles will prevail at international level. 2.1.7 Summary on assets and liabilities The result of the changing valuation culture is to make pension funds more sensitive to the risks they are running with their investments. Generally speaking, for DB pension funds, this risk is principally in holding assets other than bonds. Bonds are a low-risk investment because they are increasingly forming the basis of the liability calculation (and are the only instrument capable of securing future pension liabilities with a high degree of certainty).

14

Currency Overlay

By far the largest category of non-bond assets is equities. For several investing countries, the next largest is currencies, since several (smaller) countries have large allocations to international equities. This is followed (in decreasing importance) by property, private equity, hedge funds and other, ad hoc, assets. The latter are less important sources of risk simply because they are smaller average allocations. I will devote the remainder of this chapter to examining the role of currencies and the consequent investment risk that they create. At the end of the chapter I will also look briefly at cross-border investment by entities other than pension funds.

2.2 HISTORY OF PENSION FUND CROSS-BORDER PORTFOLIO INVESTING The countries with the two largest funded pension sectors are the US and the UK. Both have different histories of international investing 2.2.1 US Until 1980, the entire US pension sector had less than $1bn invested outside the US (or about 1 % of the roughly $390bn 1980 total in pension assets). In that year, several pension funds 4 adopted new asset allocations, driven by the desire to diversify and the ascendancy of the new quantitative modelling of risk, and the amount invested abroad quintupled to $5bn, or more than 1% of total assets. This heralded a trend which continued throughout the 1980s and much of the 1990s. By 2000, the US defined benefit pension sector had about $7000bn in total assets, of which about $800bn, or about 11%, was invested outside the US. US pension funds, with 11% in total of international assets, are low in the international league table of international diversification by percentage, but by far the largest in absolute value. By far the majority of the international assets that the US holds are equities, and it is equity portfolios, as we shall see, that form the basis for the growth of the currency overlay sector. 2.2.2 UK The growth of the UK funded pension sector paralleled that of the US. But the UK’s experience in international investing is very different, and can be divided firmly into two halves – pre-1979 and post-1979. Why 1979? Before October 1979, the UK maintained exchange controls. These had been introduced at the beginning of the Second World War as part of the panoply of measures put in place then to create, in effect, a command or ‘planned’ economy. Even though most of the legacy of regulation and control was scrapped by the middle of the 1950s, one overarching regulation remained – exchange controls. Exchange controls required that UK residents apply to the Bank of England for permission to acquire foreign currency. Permission would be given for specified purposes, including the importation of necessary foreign commodities and goods, and £50 per year (raised to £200 in the 1970s) for each individual for the purposes of foreign travel. However, permission would not be given for the purchase of overseas assets. Here the Bank of England was faced with a difficulty – at the imposition of exchange controls, there was a considerable pool of British capital invested abroad. The government decided that this could stay, but that when a UK investor wished to repatriate his investment, he could sell his foreign currency to another UK investor who wished to invest abroad.

The Problem

15

This created an arbitrary, but largely fixed, pool of overseas capital which had to be competed for by UK investors. It was called the ‘dollar’ pool, and it traded at a premium to the normal exchange rate. The premium varied according to supply and demand, and was called, unsurprisingly, the ‘dollar pool premium’. This strange corner of the financial world had one major effect on UK pension funds – they could not as a group increase their holdings of international assets, every individual increase was matched with a compensating decrease in international holdings elsewhere. Then, in 1979, in the first months of the new Conservative administration under Mrs Thatcher, exchange controls were summarily and totally abolished. The dollar pool (and premium) disappeared (incidentally causing a one-off loss for those funds already invested abroad), and from that point UK funds could invest abroad unhindered. Figure 2.1 shows the proportion of assets invested overseas by UK funds in the 20 years to end-2001. It includes equities and bonds, but overseas equities are the dominant asset class. At the end of 2001, UK pension assets were about £800bn; this puts international assets held by UK funds at about £220bn. 31.0%

Overseas as % of total assets

26.0%

21.0%

16.0%

11.0%

6.0%

81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01

Source : WM Company.

Figure 2.1 Overseas asset allocation 1981–2001 (UK pension universe weighted average)

Table 2.2 is a very rough ‘guesstimate’ based on asset values at the end of 2002, for the extent of international asset allocation. It has been compiled by the author from a range of diverse and sometimes incomplete sources, so it should be relied upon only as a general guide, not a source of definitive values. It should be noted that for the three years ending December 2002, there has been a significant decline in the size of international portfolios arising from the general decline in equity values.

2.3 CURRENCY VOLATILITY Against the background of these increasingly internationalised investment portfolios, what role does currency play?

16

Currency Overlay Table 2.2 Geographical market analysis, end-2002 estimates Approx. size of funded DB pension funds (US$bn)

Country US UK Switzerland Netherlands Japan Canada Sweden Australia Hong Kong Germany Belgium/France Total

5688 1120 298 400 643 222 71 75 15 60 76 8666

International assets %

US$bn

10% 26% 40% 25% 10% 15% 40% 30% 50% 15% 10% 14%

569 291 119 100 64 33 29 23 8 9 8 1251

Source: Record Currency Management Estimates, December 2002.

The measure I will adopt to record the movement in the currency markets is ‘currency surprise’. Currency surprise is described in detail in Chapter 3, and in Box 2.2. There is a Box 2.2

Currency surprise

Currency surprise is the difference (usually expressed in percentage terms) between the forward rate at the start of the period (say month) in question, and the spot rate one month later (or whatever the frequency being measured). It is called ‘surprise’ because whereas the appreciation or depreciation implicit in the forward rates is anticipated, any deviation from this implicit depreciation or appreciation is a ‘surprise’, and is equally likely to be up as down. The graph illustrates currency surprise: Example of currency surprise (JPY/GBP – one month)

172 170

JPY/GBP

168 One month currency surprise = GBP up by 5.63% [=(1.69831.6073)/1.6158]

166 Subsequent spot rate 164

Forward 'fall line' on 27 June

162 160 158 27 Jun

07 Jul

17 Jul Date

27 Jul

06 Aug

The Problem

17

The reason that currency surprise is attractive is because it represents the only exploitable measure of currency movements. Since an investor can buy or sell a currency forward (at the forward price), the profit or loss he will experience when that contract reaches maturity will be equal to the currency surprise. Note that to allow currency surprise to be added to or deducted from asset returns, the algebra of the calculation has the spot rate at the start as the denominator. This is a unique feature of currency surprise – all other assets have their start value as the denominator. See text for a more detailed discussion of this point. large quantity of high-quality historical currency data available. Currency surprise can be readily calculated from this data, and so we have a lot of information about currency surprise. If we take the past 20 years (1981–2001), we can illustrate the scale of currency surprise movements for major currencies in a matrix (Table 2.3). I have chosen six currencies [US dollar (USD), Japanese yen (JPY), Deutschmark/euro (DEM), pound sterling (GBP), French franc/euro (FRF)]. Note that the three-letter abbreviations in brackets are the ‘Swift’ codes (see Box 2.3) for the currencies in question – Swift codes are the currency codes internationally recognised by the banking system. I will generally use them for labels and captions, and only with introduction elsewhere. Note also that for the two largest member states of the Eurozone, I have combined the euro rates post-1999 with the individual currency rates pre-1999. Table 2.3 illustrates a number of important characteristics of long-term currency surprise: (a) Currency surprise is zero sum if taken from all base currencies – this is logically true, not just empirically true. Table 2.3 Currency surprise matrix 1981–2001 Numerator Denominator

USD

JPY 1.39% (13.15%)

USD

DEM/EUR

GBP

FRF/EUR

CHF

Average

2.03% (12.00%)

0.36% (11.67%)

0.90% 2.65% (11.91%) (13.05%)

0.62% (11.64%)

−1.04% (13.06%)

−0.50% 1.24% −0.22% (11.75%) (11.64%)

−1.64% (9.27%)

−1.11% (2.58%)

JPY

−1.42% (13.25%)

DEM/EUR

−2.02% (12.04%)

−0.63% (11.59%)

GBP

−0.40% (11.61%)

1.01% (12.93%)

1.65% (9.21%)

FRF/EUR

−0.92% (11.86%)

0.49% (11.64%)

1.12% (2.54%)

−0.55% (9.39%)

CHF

−2.65% (13.12%)

−1.26% (11.62%)

−0.63% (4.77%)

−2.26% (10.21%)

−1.73% (5.34%)

Average

−1.48%

0.20%

0.96%

−1.03%

−0.39%

1.47%

0.62% −0.96% (4.76%)

0.51% 2.28% (9.41%) (10.12%)

1.01%

1.74% (5.26%)

0.37% −1.71%

1.71%

0.0%

18

Currency Overlay

Box 2.3

Swift codes

In 1973 the international banking community formed a co-operative (‘Society for Worldwide Interbank Financial Telecommunication’ (SWIFT) to allow efficient international electronic cash transfer using common standards. SWIFT has devised unique three-letter codes for all the world’s currencies. These are becoming increasingly used not just across the market-making community, but also across the user community. They are superseding a variety of less formal abbreviations for currencies, although in common usage informal abbreviations will undoubtedly continue to have an existence. As an example, the Swift code, and the informal abbreviations, for the Swiss franc are as follows: Swift code: Informal abbreviations:

CHF SFr, SwFr, SF

Swift codes for a selection of the more common traded currencies are shown below in alphabetical order:

Country/currency

Swift codes

Country/currency

Swift codes

Country/currency

Swift codes

Argentine peso Australian dollar Brazilian real British pound Canadian dollar Chilean peso Chinese yuan Czech koruna Danish krone European euro Hong Kong dollar Hungarian forint

ARS AUD BRL GBP CAD CLP CNY CZK DKK EUR HKD HUF

Indian rupee Indonesian rupiah Israeli shekel Japanese yen Kenyan shilling Korean won Malaysian ringgit Mexican peso New Zealand dollar Norwegian krone Philippine peso Polish zloty

INR IDR ILS JPY KES KRW MYR MXN NZD NOK PHP PLN

Russian ruble (new) Saudi Arabian riyal Singapore dollar South African rand Swedish krona Swiss franc Taiwan dollar Thai baht Turkish lira US dollar Venezuelan bolivar Zimbabwe dollar

RUB SAR SGD ZAR SEK CHF TWD THB TRL USD VEB ZWD

(b) It is empirically the case that many individual currency pairs have low or zero currency surprise returns. We will come back to ‘expected return’ from currency later. (c) Annual volatilities in currencies are mostly in the 4–12% range – higher than bond volatilities (average c. 6% p.a.) but lower than equities (average c. 17% p.a.). It is also worth noting that despite the fact that the calculations are all done via logs, the returns of the same relationship turned upside down (i.e. JPY/USD and USD/JPY) are not identical. This is because we have defined currency surprise as: Currency surprise % = (Spot ratet − Forward ratet−1 )/Spot ratet−1 not Currency surprise % = (Spot ratet − Forward ratet−1 )/Forward ratet−1

The Problem

19

In the latter case, all the inverse currency surprises would be identical as long as we converted the currency surprise % to logs. The 20-year volatility averages mask very significant variations in the volatility of currencies. Popular economic history will remind most readers that there are occasional currency ‘crises’, in which one or more currencies suffer a sharp movement versus others. These cannot be called ‘crashes’, because unlike the equity market, when one currency ‘crashes’, another currency has to do the opposite of crash – it soars. We will return to this relative nature of exchange rates (one man’s gain is another man’s loss) later. We can show the variability of currency volatility by charting 20 years’ worth of 3-month moving averages of annual volatility calculated from daily prices. Figure 2.2 shows two currency pairs that illustrate this well – USD/GBP and JPY/USD. Each of the spikes are particular events which are well documented. For example, Sep 1985 in USD/GBP is the ‘Plaza Accord’ – the G5 meeting that agreed to ‘manage the dollar down’. The spike in Oct 1998 for the JPY/USD is the rapid appreciation of the yen as hedge funds unwind their ‘yen carry’3 positions.

40% JPY/USD USD/GBP

Volatility % p.a.

30%

20%

10%

0% 80

82

84

86

88

90

92

94

96

98

00

02

Year

Figure 2.2 JPY/USD and USD/GBP volatility (3-month moving average annualised volatility calculated from daily prices 1980–2002)

So to summarise, international investors have chosen to invest in international assets (mainly equities) for their diversification properties. In doing so, they wittingly or unwittingly have also invested in the currency surprise between their base currency and the currency of denomination of their equities. For the remainder of the book, whenever I refer to currency returns, I will be referring to currency surprise. 3 ‘Yen carry’ positions were those that took advantage of lower interest rates in the yen compared to the dollar to ‘borrow’ yen and invest the proceeds in dollars, and be ‘paid’ for the privilege. (The ‘borrowing’ and ‘lending’ can be replicated cheaper using forward currency contracts.) This tactic only works if the yen appreciates by less than the interest rate differential over the horizon of the trde.

20

Currency Overlay

Box 2.4

Volatility

The measure volatility as used in the financial world is a measure (normally annualised) of the standard deviation of the periodic returns of an investment, asset class or portfolio. This is the most commonly used measure of the risk of investments, and the formula for its calculation is:    n  (ri − r )2  i=1 SD = n−1 where ri = ith period return, r¯ = average value of r in the sample, and n = number of observations in the sample. If the periodic returns that are recorded are, say, monthly, √ then the convention is to annualise by multiplying the calculated monthly volatility by 12. This conversion relies for its accuracy on a mathematical theorem known as the Central Limit Theorem. The Central Limit Theorem requires certain assumptions to hold for it to accurately describe statistical behaviour, in particular that the values being measured are normal and independent random variables. Chapter 5 (and Appendix 2) describe the importance of logs in ensuring that this calculation is accurate.

2.4 CORPORATE PARALLELS IN CROSS-BORDER INVESTING 2.4.1 Foreign assets Pension funds are not the only international investors. Many companies choose to invest outside their home country, and in the past 20 years this has been on a large and growing scale. The nature of investment varies widely, but can be broadly categorised into (a) direct investment in plant, machinery and business infrastructure and (b) purchase of an existing foreign business. There are established guidelines4 laid down by national accounting standards boards, which determine the value placed on overseas assets (indeed all assets), and which also determine the method by which balance sheet value in a foreign currency is converted into balance sheet value in the home currency. However the conversion details are framed, companies’ balance sheets (and overseas earnings) have been affected directly and one-for-one by changes in the exchange rates at which they convert their overseas values into their home currency. This currency exposure has long been recognised as a financial risk to a business, and companies have employed financial and other strategies to reduce or eliminate the risk. Before we look at the economic impact of those strategies, we should look very briefly at the other side of the balance sheet – foreign debt. 2.4.2 Foreign debt As we will see below, companies with foreign assets routinely take on foreign debt. But large and/or international companies also routinely tap the international bond markets for long-term 4 These are called Generally Accepted Accounting Principles in the US, similar in the other major standard setters (UK). There is also an International Accounting Standards Board which has a remit to standardise all major standards under its auspices by 2005.

The Problem

Box 2.5

21

The euro

At the time of writing (2002), the euro is the single currency for 12 EU member states (Germany, France, Italy, Spain, Netherlands, Austria, Belgium, Finland, Portugal, Ireland, Greece, Luxembourg). The euro was launched in January 1999 with 11 members – Greece not joining initially because it failed the economic assessment. Greece subsequently joined in January 2001. Three EU member states are not members of the Eurozone – UK, Denmark and Sweden. Denmark held a referendum (in September 2000) in which the Danes voted 53% to 47% not to join the European Single Currency. The UK government has said that it will not enter the Eurozone without a referendum, and that may be held before 2005. Sweden has not held a referendum yet, but will be doing so, according to the Prime Minister, ‘before 2005’. The euro is governed by the European Central Bank, set up under EU Treaties. The ECB has the sole responsibility for monetary policy in respect of the euro. There is a common interest rate across the Eurozone, and rules limiting the fiscal flexibility of member states. The principal rule is that fiscal deficits must not exceed 3% p.a.; failure by individual states to meet this target will be met, in theory, by fines, although these have not yet been required or imposed. Portugal’s 2001 public accounts show that it has become the first state to break the 3% limit; Germany has admitted it will break the limit in 2002. debt financing at the most competitive rates. They may borrow in currencies in which they have no natural exposure. However, is it extremely unusual for companies to choose to borrow in a foreign currency on price or investor-demand grounds, and to leave such currency exposure unhedged. The instrument of choice for hedging foreign debt (or foreign bond issues) is the currency swap. 2.4.3 Economic impact of corporate currency exposure Balance sheet currency exposure is not welcomed by CFOs. They intuitively understand that it brings unrewarded volatility, even if they have not thought in detail about the rewards or the risks, nor examined the historical evidence. As a result, they expend a considerable amount of time and effort in minimising their net currency exposure. In modern reports and accounts, companies spend a significant effort in disclosing their currency exposure, the instruments they have used to reduce or eliminate that exposure, and analyses of the sensitivities of the profit and loss and balance sheet to exchange rate movements. This highlights the mismatch between the sensitivity with which companies treat their ‘own’ balance sheets, and the lack of perceived investor interest in the financial exposures of the company’s pension fund. If we make the assumption that the company is the ‘owner’ of their pension fund assets net of pension fund liabilities, and that this is symmetrical (i.e. surpluses can ultimately be recovered by the company – deficits certainly will!), then we can make an assessment on a stylised company of the relative importance of direct balance sheet foreign assets versus pension fund foreign assets. Let us take a UK company, with foreign subsidiaries accounting for 30% of gross assets, and 15% of net assets (i.e. funded with foreign currency debt). It has a pension fund equal to 100% of its net assets, and it has 30% in international assets. The

22

Currency Overlay

Box 2.6 Swaps Swaps are a class of ‘derivative’ contract. They were developed and became popular in the 1980s and 1990s to allow companies to tap bond finance in countries where there was investor demand, or to alter the maturity of the interest rate fix, again to match the different needs of both the company and investor. In a currency swap, the contract provides for one party to pay to the other a stream of interest payments, plus principal at maturity in one currency, while the other party agrees to pay a stream of interest payments plus principal at maturity in another currency. The interest rates applying to the two currencies will be different, and will reflect the domestic monetary conditions in the two countries. In an interest rate swap, the contract provides for one party to pay to the other a stream of floating interest payments (e.g. linked to LIBOR) in one currency, while the other party agrees to pay a stream of fixed interest payments in the same currency. The interest rates applying to the fixed payments, and the LIBOR-based floating payments, will reflect the monetary conditions in the currency concerned. There are many other swaps now available to convert one type of financial payment into another – the scope is almost limitless. The ‘financial engineers’ who develop and market swaps are investment banks, who will trade and hedge them on their own account. currency exposure in the pension fund of such a company (and these figures are typical for the UK) will be double that of the exposure in the company. It is safe to assume from the reporting conventions that companies currently adopt, that such a company would spend at least a page, and probably more, describing the profit and loss and balance sheet sensitivity to currency, and a similar length explaining treasury policy, instruments and risk control in relation to currency risk. It is quite likely that an effectively identical risk (currency risk in the pension fund) will not even be published, let alone analysed. This ‘immunity’ that CFOs believe they have in the pension fund may be one of the reasons that the standard remains unhedged currency exposure for pension foreign assets, but hedged in the balance sheet ‘proper’ exposures.

3 Currency Hedging 3.1 INSTRUMENTS AVAILABLE The purpose of currency hedging is to reduce or eliminate currency exposure. Currency exposure as it relates to investment assets means that the market valuation of foreign assets varies (positively) with movements of the foreign currency. It follows that an instrument whose market valuation varies negatively with that foreign currency will reduce or eliminate (if held in the right quantity) the currency variations of the asset. What instruments are there that can perform such a task? Listed here is a full (but not necessarily comprehensive) selection of the principal ones:

r Foreign debt r Forward currency contracts r Currency swaps r Currency futures contracts r Currency options Let us take these five instruments in turn, looking at three issues for each: the nature of the instrument; the costs of using the instrument; and the shape of the instrument’s sensitivity in market valuation to currency movements. 3.1.1 Foreign debt 3.1.1.1 Nature Foreign debt is fairly self-explanatory (and was briefly discussed in the previous chapter), but to recap: foreign debt will act as a hedge to a foreign asset if the proceeds from the debt are converted into the home currency (and these proceeds put on term deposit or used to buy a monetary investment that matches the term of the loan). If the proceeds of the loan are used to buy the foreign asset in the first place, then the proceeds do not need to be repatriated, since there will not need to be a purchase of foreign currency to buy the assets. The one attractive feature of using debt to hedge foreign assets is the lack of ‘interim cash flow’. That is, during the life of the debt, there will be no call on the investor for extra cash. The use of foreign debt by pension funds to hedge is extremely rare, in fact almost unheard of, because the majority of pension trust deeds do not allow borrowing. The use of foreign debt is also rare in life funds, but is much more common in investment trusts or other less heavily regulated vehicles. 3.1.1.2 Costs Using debt as a hedging instrument (without the need for the debt for gearing in the fund) is expensive. If the debt is short- to medium-term (say 1–5 years), a typical spread over LIBOR for an otherwise ungeared investment fund, acting clearing within its powers, would be in the

24

Currency Overlay

50–100 bp5 range. The equivalent deposit rate for the same entity would be at LIBID, which is 8–12 bp below LIBOR, depending on the currency. In addition, a lending institution would require documentation a couple of inches thick. This will all have a cost associated with it. It might be possible for a fund to borrow money much more cheaply (say 99.99%)

3.5 (P > 99.95%)

DEM/EUR

2.7 (P > 99.2%)

9.4.2 Risk premium The most common explanation, which is also widely used to explain the excess return of other asset classes (equities, credit), is that the FRB is a payment to investors for their accepting additional risk. The argument runs: high interest rate currencies are high interest rate because they are risky (i.e. volatile and more likely to suffer spot depreciation). Investors recognise this, and demand an expected return premium to hold these currencies – and this is the FRB. The strength of this argument is that it fits in with modern portfolio theory and investors’ risk/return trade-off. Its weaknesses, and they are serious weaknesses, are: (1) that currency rates are not dominated by international investors (trade and current account transactions have much larger volumes); (2) that in the forward market, half of the transactions are short positions not long positions, and for these investors the FRB becomes a cost, not a return; and (3) that investors in each country have different perspectives dependent on their base currency, and they will regard their home currency as riskless, even if it is a high interest rate currency.

200

Currency Overlay

Box 9.3 Currency risk premium At the core of modern portfolio theory (see Box 5.6) is the concept that investment risk is rewarded. Investors will only hold more risky investments if they expect to receive, on average, a higher return from these. Does the same hold for currencies? On the face of it, the market assigns a ‘risk’ rating to each currency in the form of the interest rate (which in turn implies a forward discount or premium in the currency). ‘Risky’ currencies (say the South African rand) have high interest rates; safe currencies (say the Japanese yen) have low interest rates. This is a pleasing description, but does it stand up to inspection? 10- Year data, USD base 14% Volatility vs. USD

JPY

12%

CHF

SEK EUR

10%

GBP

8% 6% −5%

−4%

−3%

−2%

−1%

0%

1%

2%

Interest difference vs. USD

We need to establish what ‘risky’ means for currencies. For investments, the most common measure is volatility of return. But if we apply this to currencies, from what perspective do we measure volatility? Suppose we take the US dollar as the base, we can graph the relationship between historical currency volatility historical average interest rates. The graph above makes it plain that if we compare 10-year volatility and 10-year average interest rates among five major currencies, we find a negative relationship, rather than a positive one. The same relationship holds for 5-year and 20-year data. We will clearly have to abandon the naive risk premium theory. However, we may also have to abandon the concept of volatility as the relevant measure of risk. It may be that the yen is volatile for a US investor, but the dollar is, by the same measure, volatile for the Japanese investor. On the basis of the graph above, the yen had a negative premium, but therefore by definition the dollar had a positive premium for the Japanese investor. Since all currency relationships can be inverted, we are going to get zero net risk premiums if all currency bases are taken into account. We are now left without risk premium theory for currencies. However, one final possibility is that high interest rates are seen by the market as a proxy for risky (i.e. potentially weak) currencies. The risk premium argument could be reinterpreted as claiming that high interest currencies are more likely to have positive currency surprise (i.e. the high interest rates on average more than compensate for the expected spot devaluation). This is a subset of the forward rate bias theory, which is discussed elsewhere (see Box 9.2). In summary, the expression ‘currency risk premium’ should be used with care!

Is the Currency Market Efficient?

201

9.4.3 Monetary policy and inflation A different explanation lies in the relationship between monetary policy and inflation. It runs as follows. It is governments, not markets, that set short-term domestic interest rates. They do this in pursuit of their monetary policy objectives – generally the control of inflation. Forward rates in currencies reflect these short-term interest rate differentials, and are therefore politically controlled. Governments who maintain higher-than-average interest rates in a particular period will do so to fight inflation and inflation expectations. To reduce both of these requires higher-thanequilibrium real interest rates (i.e. after adjustment for inflation expectations). This will tend to create a forward rate bias, since the subsequent movement of the spot rate will reflect the actual (lower) inflation. The same effect will also happen in reverse – countries pursuing low interest rate policies do so to stimulate demand, potentially raising inflation above expectations. The weakness in this explanation is that the relationship between short-term interest rates, spot rate movements and inflation is highly unreliable, and a poor basis for a stable phenomenon like the FRB.

9.4.4 Nominal rate illusion An explanation which does not suffer from the same shortcoming as the two above is the nominal rate illusion argument, which runs as follows. Markets are dominated by the behaviour of their players, and therefore by their perceptions and expectations. Currency markets are reported and traded in spot rate terms, and ‘currency surprise’ is a little known concept. However, all profits and losses in the forward market in currency arise out of currency surprise, which, to recap, is the result of currencies not ending up where their forward rate predicts they will. This dissonance between profits and losses actually experienced (currency surprise), and the direction of the market and its ‘pressure points’ as actually reported and understood, means that FRB can flourish. How is it that this is not arbitraged away by the force of the market recognising the profit opportunity? Currencies are not principally investment instruments, and the vast majority of players in the currency markets are not ‘investing’ in currencies. This means that not just the perception, but the reality, is different for different players. Currency exchange rates have always attracted the attention of central banks. They are often charged with defending particular (spot) exchange rates, and will ‘smooth’ extreme (spot) movements by intervention. They are not profit centres, and therefore they have a political imperative to perpetuate the ‘nominal price’ illusion by their behaviour. A significant proportion of market participants are corporations, and many of these will have target exchange rates expressed in spot terms, and the movement of the spot rate will be a critical determinant of their currency transaction behaviour. This emphasis on spot rates is extremely widespread. Major commodities are priced around the world by conversion of their ‘core’ currency (mainly dollars) into the users’ currency at the spot rate. When a German oil company is buying Rotterdam crude in dollars, it is the spot euro/$ that determines the oil price in euros – and ultimately the price of petrol at the pumps in Germany and the rest of Europe. Similarly, when a company has multiple production facilities in different currency blocs, it is the spot rate that is the determinant of the lowest cost producer. To a great extent for corporations, the nominal price illusion is not an illusion – nominal prices are hard reality. Corporations inhabit the ‘current account’ world, and they have to respond to immediate

202

Currency Overlay

opportunities and threats arising from exchange rate movements, not to notional accumulation of ‘currency surprise’ (which calculations are irrelevant). But the scale and importance of the ‘irrelevant’ calculations is significant. As an example, the yen was on a downward path in 1995 (from a high of 82 yen/$), and in October crossed the 101 yen/$ level and continued down (i.e. the number rising). It reached a low of 144 in July 1998, and then promptly recovered its poise, to rise all the way to 101 in December 1999. It then turned round and fell again. It looks at face value as though the 101 in 1999 was the ‘same level’ as the 101 in 1995. From the point of view of the ‘exchange rate charts’, the market-watchers, the competitiveness of Japanese goods, the price of oil in yen and the opportunities for exporters to sell to Japan, this is true. But the JPY/USD currency surprise index which was 101 in October 1995 was 126 in December 1999. So some 25% of yen weakness in the currency surprise was concealed in the nominal price illusion, and any market player who maintained a short position in yen throughout this period (exploiting the forward rate bias) would have made a 25% cumulative profit, as opposed to nothing measured by the spot rate. 9.4.5 Other inefficiencies I have demonstrated the evidence for, and some possible explanations for, two well-established and widely accepted inefficiencies. There are more inefficiencies, although perhaps none as readily demonstrable and as transparent as trends and the forward rate bias. Active currency overlay managers have successfully exploited more inefficiencies than just trends and the forward rate bias over the past 10 years, and have as a manager group added value (see Chapter 11). This observation itself is unusual and needs elaborating.

9.5 A SUCCESSFUL UNIVERSE? How can a universe of managers add value versus an objective benchmark? In a market where the universe of managers in a survey constitutes all the managers in the market, then it is logically impossible for the universe to outperform the index (which is the objective benchmark), since the universe being tested is itself the complete constituents. Since managers incur costs and indices don’t, then managers as a group will not just not outperform, they must underperform. All the available studies of active management that I have seen, where the active manager universe is a significant proportion of the market (which is most equity and bond markets), illustrate the reality of this finding: that active managers as a class do not (and cannot) add value. The best analogy for this comes from sport. While the absolute skill in the UK Soccer Premiership6 is undoubtedly very high, the inescapable fact is that the average final position of the teams in this league is 10.5. The interesting question then becomes ‘are there any individual teams which have a consistent record of success’ (i.e. consistently higher than 10.5th place)? In the active manager context, this is a question about individual management firms (or possibly even individuals within firms) – not about the universe. The average score of the universe is always 10.5.7 6

The top UK soccer league, which has a fixed membership of 20 teams. The analogy can neatly be extended to survivor bias as well. Since the lowest three teams are relegated each season, the average 5-year premiership position of teams currently in the premiership must be a lower (i.e. better) number than 10.5 (and cannot be higher) as some teams will have less than a full 5-year record. 7

Is the Currency Market Efficient?

203

What is different about currency? From the foregoing sections on the structure of the currency markets it is clear that the active money available to exploit the persistent inefficiency identified above, and any others, is inadequate to fully eliminate them. On very rough guesstimates based on the volume of cross-border assets, and the constraints on asset managers, I estimate that the total of long-term active money directly available for this purpose is less than 1% of FX market turnover.8 With this very small active, profit-seeking component, it is entirely consistent with logic that an identifiable sector of active managers (currency overlay managers), who all exploit one or more persistent inefficiencies, can outperform an objective benchmark or market index. The corollary of this is that the remaining 99% non-profit-seeking sector must lose money against the same benchmark. Why don’t they notice or mind? They don’t notice for three reasons. Firstly, the amounts involved are tiny, and get ‘lost in the wash’. Secondly, many participants have to trade in the FX market to cover international payments, so they do not have the choice to withdraw even if they thought they were losing an unidentifiably small amount in their trading activities. Finally, the measurement of performance in FX is fundamentally dependent on the customers’ perception of the benchmark, and their base currency. This means two participants can simultaneously think they are making money. Section 9.5.1 gives a simple example of this phenomenon. 9.5.1 An example of different perspectives in the FX market Customer 1 buys $1m dollars with euro from the bank, and simultaneously another customer (customer 2) sells $1m to the bank for euro. The bank charges 0.05% difference between the two prices, and it is profitable (and happy). Suppose the market rates were (in $/EUR) 0.9000 and 0.9004 respectively. Let us suppose the rate then moves to 0.9500 over the next few weeks. Customer 1 is a euro-based importer, and is delighted that his imports for next month and subsequently will be cheaper. He plans to reduce his prices (which will stimulate demand) and simultaneously increase profits because of the 5% input price reduction. He may take more forward currency cover at these levels to lock these new rates in. Customer 2, a US-based investor, congratulates himself on the timing of his unhedged euro investment – it has already gone up 5% in value. In these complex and varied circumstances, it is easy to see that very small and unidentifiable amounts of ‘losses’ will get lost.

9.6 SUMMARY ON EVIDENCE FOR INEFFICIENCY 9.6.1 Weak form efficiency I have presented here two independent pieces of evidence that give very strong support for the contention that currency markets are not even weakly efficient. Both of these pieces of evidence refute the weak form of the efficiency model, namely that prices are not predictable from historical prices or price patterns. If weak form efficiency is refuted, does this mean that semi-strong and strong form efficiency are refuted as well? This is an interesting question, to my knowledge not examined widely in the literature. 8 Currency overlay = $110bn; cross-border portfolio assets = $1.5trn; 10% asset manager hedging leeway = $150bn. Hedge funds = $100bn of risk. Total active = $360bn. T/O 3× value p.a. Therefore long-term active turnover = $1.08trn p.a. Total FX turnover = 260 × $1.2trn × 41% (customer proportion) = $128trn p.a. Therefore active proportion = 1.08/128 = 0.84%. All 2000 estimated values, various sources.

204

Currency Overlay

9.6.2 Semi-strong form efficiency This form of efficiency states that all public non-price information is already in the price, but private information is not. To refute this form requires evidence that there is public non-price information not in the price. This in turn requires evidence that non-price-based models (i.e. fundamental models) can be constructed which give us predictions about future prices in which we can have (statistical) confidence. So this boils down to the question ‘do fundamental models or forecasting work?’. We will explore this question further in Chapter 10. 9.6.3 Strong form efficiency In the mainstream developed currency markets (which includes probably 15 currencies, and therefore 105 pairs [=(15 × 14)/2]), there is virtually no availability (and little concept) of private information. Currencies are not like equities or other corporate securities: they are not ‘owned and controlled’ by any one individual or group of individuals. Whereas a disappointing trading statement, composed by the CEO and CFO of a quoted company, will almost certainly cause a fall in the price of the company’s share price, a piece of ‘bad news’ about an economy may not necessarily cause a fall in the exchange rate. For example, ‘weak demand’ may be interpreted by the markets as reducing the possibility of a rate rise (probably bad for the exchange rate), or as reducing the demand for imports (probably good for the exchange rate). In addition, the length of time that price-sensitive information remains in the private domain is very limited. Most statistics are published very quickly after they emerge from compilation – often within 24 hours. This combined with the paucity of such information remaining in the private domain makes strong form efficiency in the currency markets hard either to prove or disprove – in fact probably irrelevant. 9.6.4 Transactional efficiency As mentioned above, all of this makes no statement about the transactional efficiency or otherwise of the currency market. The evidence presented elsewhere in this book is strongly supportive of the argument that the currency market is extremely efficient from the perspective of transactions and liquidity. The evidence presented shows that the depth of liquidity is the deepest, and the cost of transactions is the lowest, of any of the global markets. This low cost of transactions does not sit at all uncomfortably with the informational inefficiencies identified in this chapter. In fact, as one consultant specialising in the currency market says: ‘wouldn’t you expect active managers to be able to make money if over 90% of assets were indexed, transactions costs were 5 bp or less, there were no taxes, duties, or other frictional costs, and daily turnover was over $1trn per day?’

10 Active Currency Overlay – Management Styles In this chapter I will look at two main issues – how active currency managers come to design the active processes that they do, and what those processes are. To use an architectural analogy, this is the same thing as firstly looking at how architectural styles developed (the cultural, technological and artistic background) and therefore why an architect designed in the way that he did. Secondly, again using the architectural analogy, what the styles are (i.e. a description of classical, gothic, modernist, etc.).

10.1 THE PROBLEM The problem that all active currency managers (or potential active currency managers) face is how to use information available today to predict with a measure of success (at least a measure Box 10.1

Cross hedging

Cross hedging is where a currency overlay manager undertakes a forward contract or hedge between two currencies, neither of which is the home currency of the investor. Such activity either passes risk from one currency to another, or creates new currency risk; it does not by itself hedge or reduce it. Under what circumstances is cross hedging likely to be used? Firstly, there are circumstances in which the owned (i.e. the target of the underlying investment) currency is relatively illiquid and expensive to trade, but closely associated, either formally or informally, with another foreign currency. An example for a European investor might be a Latin American currency, with the US dollar being the associated currency. It would be quite possible to passively hedge the target currency back to US dollars (i.e. cross hedge), and then run an active overlay between the US dollar and the home currency. Secondly, the overlay manager may wish to exercise more freedom to choose the currency pairs on which he is likely to add value. With the client’s agreement, he may release the restrictions on (i) the client’s base currency always being one half of the traded currency pair; and (ii) the weights of the active bets being determined by the underlying asset allocation. In theory cross hedging could have any number of legs. (Currency A hedged to B; B hedged to C; C hedged to D; D hedged to home currency, etc.). However, active cross hedging is a form of gearing. It allows more active bets on the same asset base; it will engender a higher tracking error, higher expected value added and a higher potential downside. There are few circumstances in which cross hedging would be a valid strategy for passive hedging. Only if the currency market was quoting irrationally (bid/offer spreads for currencies A:C being wider than the sum of bid/offer spreads for currencies A:B and B:C) would passive hedging have a financial incentive.

206

Currency Overlay

more than random) what will happen to exchange rates in the future. This is a problem simply stated, but extremely difficult to solve. I should state here that there will be a group of readers who will believe that the problem does not have a solution. Such a belief will be founded in their own experience, and based perhaps also on what they see as inconclusive evidence to the contrary. My goal is to present the relevant information clearly and with insight to both sceptics and believers. I will also declare my beliefs so that the reader can gauge the position from which I am speaking: The evidence from over 20 years’ experience of analysing and trading in the foreign exchange markets has convinced me that highly sophisticated managers can consistently beat the market (and therefore add value) over time. However, doing so is difficult (much more difficult than it might appear), requires time and patience as well as expertise, and will be a great deal more anticipated than achieved by those managers who try.

It is worth stating at this point that most active overlay mandates will constrain the active manager to vary hedge ratios between 0% and 100%, irrespective of the benchmark. It is within this constraint that the following discussion takes place.

10.2 MODELLING AND FORECASTING Faced with the problem above, almost all currency managers turn to historical data, and quantitative modelling, to assist them in unravelling the mysteries of the currency market. If they do not do so, what is their guide? They may have some anecdotal or period-specific experience of a particular currency move or moves, but without mathematical support, any views or beliefs they may have are untested and unverifiable. So how do they go about modelling? 10.2.1 Modelling – Occam’s razor1 Many professionals in currency management, particularly in stand-alone currency overlay, use mathematical models to analyse and exploit historical inefficiencies in price movements, and to determine (more or less formally) their active positions for their clients. The advantage of models is that they allow scientific methodology to be applied to the process of active management. A hypothesis is established – say that currencies have medium-term trends, or that volatility in a period is positively influenced by a decay in volatility in previous periods. A model is built to test the hypothesis or hypotheses, and the model is tested on historical data. Modern computers and statistical techniques are now very sharp tools for conducting this analysis, and it is not difficult to get statistically significant results out of this process. Statistical significance means abandoning the efficient market hypothesis, and accepting that the chosen model has explanatory power. However, there are significant pitfalls in applying naive modelling processes. The first one is that a competent modeller can get very high levels of explanatory power modelling many data series.2 Increasing model complexity, and relaxing the ‘intuitive tests’ that good models satisfy, can allow explanatory power to rise almost without limit, to the point where information 1 Occam’s razor states that where there is a choice between two equally valid explanations or approaches to a problem, you should choose the simplest. William of Ockham (Occam) (1285–1349) was a medieval English philosopher. Einstein had a similar view – ‘theories should be as simple as possible, but no simpler’. 2 This also unfortunately applies to data series which have no genuine explanatory variables. It is frequently possible to get significant explanatory variables by modelling randomly generated series with enough different ‘explanatory’ series.

Active Currency Overlay – Management Styles

207

ratios3 apparently exceed 1. Increasing model complexity increases the historical fit, but the evidence is that it also increases the gap between history and the future: as the explanatory power increases, the relative predictive power decreases. The intuitive explanation for this is as follows. Historical price patterns in history are the distillation of the almost infinite variety of human behaviour, economic conditions and chance. When presented with a price series, and other economic variables (relative interest rates, inflation rates, economic data, flow data), a modeller will design a mathematical model to ‘explain’ the movements. The model may have a variety of intuitive bases, but can only capture, at best, a tiny subset of the actual forces which ‘really’ applied to the historical price series. There will be a huge amount of unexplained price movement. The temptation will be to try to ‘explain’ the unexplained movement, and modern modelling techniques will allow the modeller to do this by adding autoregressive terms, non-linear functions, and almost limitless exogenous variables. The explanatory power can only go up when this complexity is added. But unless this added complexity is genuinely capturing underlying behaviour, the added complexity is just ‘curve fitting’ – that is it ‘explains’ the historical movement by nature of identifying particular correlations and relationships that are supported only by the data – not by any ex ante intuitive beliefs. It is only a short step from this to ex post rationalisation of the chosen model, and we have the makings of a very unsuccessful model. In reality, there is a continually varying combination of generators of currency returns which include long-term structural elements, and shorter term factors which will create a great deal of noise if ignored, but by contrast will lead to disappointment and failure if modelled. The yen hedge fund debacle of 1998 was a classic case of a large amount of money temporarily behaving in a particular way, and thereby moving the market. Modellers were faced with a dilemma (which faces them less obviously all the time): either this behaviour creates a lot of ‘errors’ in models which ignored this factor, or it is captured and ‘explained’ by new complex model features. Unfortunately, the causative factors for this behaviour, namely a herd play on the ‘yen carry’, evaporated in a few days in October 1998 with a massive fall-out for the hedge fund players, and a failure of the new ‘explaining’ elements of the model. This feature of model building is very difficult to guard against in inexperienced hands, since enthusiastic exponents will always be looking for ‘best fit’ models, and will be under commercial pressure to present them with short (or ‘theoretical’) track records. It takes a very experienced modeller to reject models with good fit and good ‘stats’ (split-test results, etc.) because he does not believe them to be capturing a stable non-random characteristic of the market. With all these caveats, is there a role for modelling in active management? 10.2.2 Can models work? The answer is yes they can – but it doesn’t mean they all do. Models of physical processes are the life-blood of technological advance, and many model the real world near perfectly. However, as the explanatory power of models decreases with increasing ‘random’ elements, their validity is increasingly hard to be sure about. You only need to think about weather forecasting to understand some of the problems. 3 Information ratio = annual value added %/annualised tracking error %. Annualised tracking error is the annualised standard deviation of the differences between manager performance and benchmark performance. Successful active managers (currency or any other asset class) tend to deliver live information ratios in the 0–0.5 range. It is rare (although not impossible) to find long live track records (which haven’t been ‘cherry picked’) with information ratios higher than this.

208

Currency Overlay

Models of financial market behaviour are bound to have large unexplained elements, and they also have a further, logical problem already referred to above. In most financial markets (e.g. equities and bonds), if a model is used to observe a market inefficiency (i.e. explain a proportion of the returns), and then used to exploit it, the very act of exploiting it will (if done in sufficiently large volume) change the market’s behaviour and thereby undermine the model. However ‘reflexive’ and ‘clever’ a player – perhaps calculating the extent of a particular model’s use and predicting how its use will subsequently affect the markets – nevertheless in all markets in which objective benchmarks can be established, one player’s outperformance will be another’s underperformance.4 It is quite hard to imagine a model sufficiently reflexive and ‘game theory’ orientated to be reliably successful in such an environment. Hence the relatively small proportion of purely quantitative houses in active management of equities and bonds. Box 10.2 Alpha The expression ‘alpha’ means ‘value added’ or ‘excess return’ over and above the market return available for that asset class. The use of the greek letter α arises from an equation in the Capital Asset Pricing Model (CAPM) that describes the pricing of an actual portfolio in relation to its riskiness, namely: Portfolio return = α + β(market return) + ε where ‘return’ is the return in excess of the risk-free rate, β is the elasticity of the portfolio return with respect to market return, and ε is a random error term with a zero mean. Put into words, alpha is the additional return in a portfolio over and above the return expected from a passively managed portfolio of a certain riskiness. The expression ‘alpha’ is often used as shorthand for the ‘skill’ of a manager. The expression ‘alpha’ was popularised by Michael Jensen, then a Chicago PhD student, who studied (in ‘The performance of mutual funds in the period 1945–64’, Journal of Finance, 1968) the performance of actively managed mutual funds compared to the performance of market indices. He suggested that ‘alpha’ was an effective measure of stock-picking ability in such mutual managers, although he found a disappointingly small proportion (a minority) of his subjects did outperform the market index. In the context of currency overlay, ‘alpha’ is universally used to mean the value added of the currency manager versus the benchmark. Since there is no concept of ‘beta’ in currency management, alpha is a measure of skill. It does not, however, say anything about the nonsystematic risk that a manager is taking to achieve his alpha – this is commonly measured by the information ratio. What about active currency models? As described above, there is a very large, active (in the sense that they are dealing), but non-profit-seeking customer sector in the currency market. If this sector behaves in a stable way (because of its sheer size and diversity, and because it doesn’t have a choice about a lot of its deals), then in logic it should be possible to build a model which 4 This is only true if all the players in the market are represented in performance measurement, and the benchmark represents the whole relevant universe of investments. Under these conditions, the universe must underperform the benchmark – because the former have costs and the latter does not. However this does not imply that active management cannot make money – just not all active managers.

Active Currency Overlay – Management Styles

209

exploits the inefficiencies which this behaviour throws up without the model being self-limiting or self-destroying. However, stable features are likely to be few and rather general (trending market, forward rate bias), so such a model will have to successfully contend with a great deal of noise. The level of noise may make the process of creating a successful track record quite long-drawn-out, but if the model is genuinely exploiting a stable market feature, which is not submerged by transaction costs, then it will slowly haul ahead of a passive benchmark. Section 9.2 tells us that this can happen at active manager universe level as well. Box 10.3

‘Star’ investors

‘Star’ investors are a modern, and interesting, phenomenon. They exist in many active investment management firms, and they are believed by clients and their employers alike to possess exceptional investment skills which translate into exceptional investment performance. How far are these perceptions based on reality, and how much are they created and promoted by the investment firm and the stars? There are two obvious tests to apply to answer these questions. The first is to ascertain whether the firms in question are able, with any consistency or reliability, to beat the benchmarks against which they are measured. The second is whether this performance is portable with respect to the stars – that is, if they move firm does their performance move with them? Both of these questions are hard to answer, but the first is easier. Most active equity managers do not have records of consistent outperformance. There is an active debate whether US equity managers are, as a group, so sophisticated and so dominant in the US equity market that all inefficiencies have been removed. This would both undermine the case for the ‘star’ system and for active managers in general. American investors would argue that active outperformance is still possible in the ‘less sophisticated’ foreign equity markets, although such evidence as there is appears to point to country (and currency) selection as being the greatest contributors to outperformance. The evidence for ‘portability’ of performance is almost impossible to establish. Firms do not provide performance by portfolio manager; much of the support for the ‘star’ system is therefore anecdotal, or explainable by chance and survivor bias. Hedge funds are a potential source of evidence in this area, since they are often linked to (or founded by) particular investment stars. However, the hedge funds with long track records are so few, and the benchmarks against which to measure them so debatable, that it is impossible to abandon the hypothesis that active equity outperformance is the result of chance. In currency overlay management the star system does not really exist. Most overlay managers have adopted highly quantitative approaches to active management, and it is the reputation of the quality of the analysis and modelling process, rather than reliance on a ‘star’, that is the key distinguishing feature of overlay managers.

10.2.3 What about active management without models? Successful active managers are like successful chess players. They need to be clever, knowledgeable and resourceful, and be able to respond to new situations in an effective way, and ultimately, they need to beat their peers. It is therefore likely in many markets that personal knowledge (knowledge of target companies, knowledge of market behaviour) is the best route to beating the other market participants. However, even a game-theory-driven game like chess

210

Currency Overlay

can be codified. With modern computers, all but the very best players are now being routinely beaten by computer models. Even the best player in the world was beaten by a computer model in the celebrated 1997 series.5 Very good players, in markets as much as in chess, operate in a non-random way which could in theory be codified into a model. Arguably the most successful living investor, Warren Buffet, has publicly set out his criteria for investment in the stock of a company. He has (largely) stuck to his ‘model’ despite it looking poor in some circumstances (e.g. the Internet bubble). The criteria are simple, and in theory easy to copy. They have not been widely copied for many reasons, the most pressing being that those asset managers with enough money under management to be able to take strategic stakes in top-of-sector companies are required by their clients to deliver index-like returns. Warren Buffet-like investing is inconsistent with index-like returns. Models only codify consistent ways of behaving, however complex the thinking and the input data. In the currency market ‘personal’ styles of active management have been less popular and successful than more quantitative approaches because the levels of noise are high, the horizons long, the quantity of data high, the ‘inside information’ scarce or non-existent, and the ‘reflexive’ element quite low. The latter is a consequence of the huge size and diversity of the market. However, the previous section explained that the most likely long-term winners in active currency will have relatively simple models, and these do not necessarily need to be formally codified as long as their rules are consistently applied. 10.2.4 Dealing and practical execution The best models in the world are no good unless an active currency manager has a well-designed process for converting model output (or indeed discretionary judgement) into live deals. This is no easy task, the difficulty of which is much underestimated. Let us suppose that a manager is developing or refining a model. The inputs to the model will include the current exchange rate (spot and forward), a variety of historical rates (spot and forward), and an array of other, exogenous variables. To be effectively tested, the testing procedure must exactly match the live procedure to be adopted once the model is live. Otherwise it’s like testing a new car on a smooth, flat, test-track, and the customer taking it out on real (rough and hilly) roads, expecting it to perform to test-bed specification, and being disappointed. 10.2.5 Timeliness of inputs For all modellers with an eye to forecasting, a firm understanding of the timeliness of inputs is critical. Let us suppose that one of the exogenous input variables is (say) balance of payments data. This data may be released regularly at, say, 10.00am local time once a month. Whatever use the model makes of this data, it must always ensure that it uses it (and only contemporarily available information) after this date and time. If the announcement also includes revisions of past data (as is common), then only those contemporary revisions are capable of being used. It is common for economic data to have a series of revisions; and unless each of 5

Kasparov vs. IBM’s Deep Blue.

Active Currency Overlay – Management Styles

211

these is independently stored as the latest information on a particular date, then the model is in danger of targeting post-dated data. If only (finally) revised data is stored in one time series, then the model will be anachronistic – using data at time t that was only available at time t + 1. Similar problems occur with price data. If the technical component of an active currency management model uses contemporary prices, exactly what prices are these, and how are they reflected in the live execution process? Let us suppose a technical model is designed with, and runs on, daily prices. It assumes (not unreasonably) that it can execute deals at the stored historical daily price, with an adjustment for bid/offer spreads. Does the execution process only run on daily prices, or does the dealer have a price-level signal stored in the computer ready to deal at any time once the level is breached? If the latter, then this is a significant deviation from the model, and either the model should be revised to use higher-frequency prices, or the execution should be amended to deal at only once-daily price ‘fixings’. Maintaining a gap in execution process between the model and live dealing significantly reduces the quality of the model, because the model has no information about higher frequency data. 10.2.6 Judgement and modelling The exercise of discretion in the execution of a quantitative model in the currency market sits very uneasily with the concept of the scientific method in modelling. Let us suppose a model generates a signal. The signal is passed to the portfolio manager, who then exercises his judgement as to the timing and manner of the execution (and indeed the advisability of the deal at all). The question that needs to be answered of such a process is this: does the discretion of the portfolio manager improve the operation of the process?6 To answer this, the firm will need accurate record-keeping of the tradeable prices at the time the model’s signal is issued, and of the final (judgement-influenced) deal. If this data is available, then the portfolio manager’s input can be analysed, and value-added and risk measures applied to determine its success or otherwise. The model’s live track record (which is the non-judgementinfluenced trades) can be analysed and compared to the simulations on which the model was constructed. Quite separately, the portfolio manager’s deals (the gap between the model’s deals and actual execution) can be analysed. If the portfolio manager is consistently successful, then he is a generator of currency ‘alpha’,7 and not dependent on being ‘given’ trades by the model. It may also be worth the firm trying to understand what (further) inefficiency the portfolio manager is exploiting (not being fully exploited by the model) that allows this value added. Finally, the firm might consider whether this skilled input from the portfolio manager can be codified, and itself turned into a model. If the portfolio manager is not consistently successful, then there might be good reason to eliminate this source of uncertainty from the firm’s process. If the data for this ‘alpha attribution’ is not available to the firm (or the client), how can either be sure that the discretionary exercise by the portfolio manager is worth doing? Clients who get the opportunity should ask a firm with discretionary override to justify its policy with data. 6 This section refers to value-added processes in currency markets, but not necessarily to other, less liquid markets. Because of the liquidity of currency markets, model-generated deals do not need to be ‘finessed’ to get good execution. The quoted price is in most instances instantaneously tradeable at narrower than quoted spreads. See Chapter 4 for more on this. 7 ‘Alpha’ is commonly used to denote ‘value added’ versus a benchmark.

212

Currency Overlay

10.2.7 Deal execution Currency managers will be judged on their ‘after-dealing-costs’ performance. While this should prove a strong spur to ensure that their deals do not attract excessive bid/offer spreads from the foreign exchange market, there may be some factors which influence an overlay manager’s choice of counterparty for his client. The most obvious factor (and potential conflict) is where an overlay manager is owned by a bank with an active FX dealing room. Deals undertaken by the overlay manager with his parent will always create an element of uncertainty – and probably the most transparent solution is for such a manager to forbid themselves deals with their parent. Box 10.4 Active currency styles Currency overlay managers are commonly categorised into three styles: fundamental, technical and dynamic. There is a fourth, option-based, which in many ways (but not all) is a different manifestation of dynamic hedging. Fundamental managers believe they can exploit price inefficiencies using models and processes in which economic and financial data is used as the ‘exogenous’ variables. Examples of such data include balance of payments, capital flows, relative price levels, relative monetary conditions, etc. Fundamental managers have to accept that there may be quite long lags (measured in years not months) where their models may not work effectively, and this has reduced to a handful the number of overlay managers who use exclusively fundamental techniques. Technical managers are philosophically the ‘opposite’ of fundamental managers. They tend to ignore completely external economic variables, and argue that price and price history provide the most effective mechanism for exploiting inefficiencies. Their typical approach would be to model price history to determine successful trading rules – say like buying a currency when its 5-day moving average price crosses its 25-day moving average. Good technical managers in currency overlay have added value over time. Dynamic managers are a group that aims to create an asymmetric return – running profits and cutting losses. Many dynamic managers use currency option technology to do this. Most use forward contracts for this process, but a smaller group use option instruments (see below). Option-based managers’ approach is to exploit what they see as systematic differences between implied and actual future volatility. In many ways the inefficiencies they exploit, and the pattern of excess returns, are similar to dynamic managers’. There is a live debate as to whether these categories are accurate or appropriate. Few of the overlay managers active at the time of writing would fully subscribe to these descriptions.

10.3 ACTIVE MANAGEMENT STYLES I want to turn next to the common styles in active currency management. The naming of styles in asset management is always a little contentious – each style is named by its proponents in such a way to make it attractive or ‘natural’. Two popular styles in active equity management are ‘growth’ and ‘value’. Both sound good. But they are simply two halves of the spectrum of one particular value or group of values – essentially price/earnings ratios. High ratio = growth

Active Currency Overlay – Management Styles

213

stock; low ratio = value stock. They can’t both simultaneously outperform the combination of the two (which is the whole market). Active management styles are in reality a reflection of the types of model and model input that a manager chooses to employ. Once a manager chooses to accept a particular relationship as having predictive power, it will determine his ‘style’ as seen by the outside world. The most common categories in active currency overlay (see Box 10.4) are as follows:

r Fundamental r Technical r Dynamic r Option-based I will take each in turn, although I will look at option-based styles before dynamic hedging, since this gives the introduction to option-like payoffs that dynamic hedging is based on. 10.3.1 Fundamental Fundamental managers believe that currency markets are not semi-strong efficient. That is, they believe there is non-currency-price public information that they can (regularly) obtain that will give them greater than 50% success rate in their exchange rate positions at some horizon in the future – i.e. make their clients a profit. Fundamental forecasters are deluged with possible contenders for explanatory data – inflation, interest rates, money supply, GDP growth, consumer spending, central bank reserves, bond yields, cross-border asset flows, balance of payments, foreign direct investment, central bank intervention policy, government borrowing, finances and debt, etc. The list goes on and on. I could list virtually every category in each government’s statistical output of economic data, and hardly any would be completely irrelevant. How do managers sift this data for relevance, and how do they build effective models? The sifting of data is generally determined by the hypotheses that a manager wants to test. If a manager does not have a clear ex ante hypothesis, then that is a reliable signal that the model is going to end up with a strong ‘data-mining’ element.8 Let me give an example of how the testing process might work. (I have kept it simple to illustrate the process, not to build a credible model.) 10.3.1.1 Hypothesis Let us suppose that the hypothesis is that exchange rates that are higher than purchasing power parity levels have a tendency to go down (increasing with the distance from parity levels), and that rates that are lower than purchasing power parity levels have a tendency to go up (increasing with the distance from parity levels). It is intuitively appealing and the inputs are relatively simple. However, it is not quite as simple as it may seem. This hypothesis may be true, but still not be a way for the manager to make money. This is because the manager can only deal in interestrate-adjusted spot rates – i.e. currency surprise. So for the model to be any use, the hypothesis must be changed to: ‘spot exchange rates that are higher than purchasing power parity levels 8 Data mining is the (dismissive) name given to models which are built by choosing as explanatory variables those series that fit the data, rather than those which reflect a well-defined theory. It is pretty much the same thing as ‘curve fitting’.

214

Currency Overlay

Box 10.5 Capital account National accounts’ economists separate balance of payments data into current and capital account. The way the national accounts process works is that physical imports and exports are recorded in detail, while imports and exports of services (including interest and dividends in both directions) are much more roughly estimated (since there is no port documentation) as ‘invisibles’. If these two do not add up to zero, then the balancing item is called ‘capital account’. Interestingly, countries with exchange control regulations may have much better invisibles data than those that don’t, although exchange controls will almost certainly also spawn a black market, where there is by definition no data. Capital account transactions, being a residual, are enormously varied. Taking, say, the US perspective, the most obvious ones are cross-border lending and investment in securities – e.g. Japanese public and private sector purchase of US Treasury and other securities ‘financing the US deficit’. Other less obvious forms of capital account transactions are, for example, US investment institutions hedging their foreign exchange exposure or a quoted US company building new US plant with the proceeds of a share placing. All these are examples of capital inflows into the US. The last case (where I am assuming some foreign ownership of the quoted company) is that foreign owners subscribe to the share placing. The current/capital split is defended on the grounds that current account performance is a great deal more predictable than capital account. But it is worth remembering that the sum of the current and capital account is always zero; which implies to the logical noneconomist that the capital account is just as predictable as the current account. The only question is the exchange rate that is necessary to achieve that. have a tendency to generate negative currency surprises (increasing with the distance from parity levels), and spot rates that are lower than purchasing power parity levels have a tendency to generate positive currency surprises (increasing with the distance from parity levels)’. If this hypothesis proves to be supported by the evidence, and is stable over time, then in theory it could be the basis for a fundamental model. 10.3.1.2 Inputs 1. We can proxy the changes in purchasing power with the relative change in consumer price indices in the relevant countries (we will need the date that the later of the two inflation levels is announced to avoid the anachronistic mistake). 2. We need a level for purchasing power parity. This could perhaps be the inflation-adjusted average spot rate over a long data period (say 20 or 30 years). But see below the discussion of purchasing power parity. 3. We will have to have a series of spot rates to calculate where current rates are in relation to purchasing power – and forward rates have no role in this calculation. 4. We will need a series of currency surprise – or more particularly an interest-rate-adjusted index of the exchange rates. We need the latter because we might want to vary the horizon over which this hypothesised effect applies. It might not work in a month, but might work over two years.

Active Currency Overlay – Management Styles

215

We also need to pick a currency pair to start with. We will use USD/GBP as an example, but in a live modelling environment, this would be many pairs. We need to choose a modelling frequency: most econometric models are no more frequent than monthly, whereas trading models are almost always daily frequency. 10.3.1.3 Regression Most econometric models use a technique called ‘regression’, whose basic principles are as follows. A hypothesis is established, say that future changes in log currency surprise are determined in part by the distance that the spot rate is from purchasing power parity. If interestrate-adjusted log currency index = ct , then future changes at horizon n months = ct+n − ct . Suppose that xt = % over/undervaluation of sterling using PPP values as the benchmark. The regression equation yt = a + bxt + e where yt = ct+n − ct is a standard linear regression equation, testing whether y is explained by x; a and b are constants (which will be calculated in the regression), and e is the error term (i.e. the unexplained variations in y). 10.3.1.4 Correlation Because we are only testing the relationship between one external factor (PPP) and changes in future exchange rates, we can run correlations between these two series. We will have about 20 years of monthly data, so about 240 points in the series. For this number of observations, correlation coefficients above about 0.13 (or −0.13) are statistically significant (i.e. we can reject the hypothesis that the two series are unrelated). 10.3.1.5 Trading model As well as testing the statistical relationship between series, we can also test whether trading rules set up to attempt to exploit the relationship can make money. Any trading will be conducted in the currency surprise series (since this is exploitable), not the spot return series (since this is not investible). The rules can be infinitely varied. I have shown three. The first two take positions for certain predetermined horizons (say 24 months): they are a variable position (i.e. proportionately shorter of sterling as it gets increasingly overvalued) and a ‘long/short’ switch as the spot rate crosses PPP. The third model does not have a horizon, instead it takes proportionate positions each month dependent on sterling over/undervaluation vs. PPP. These are adjusted end-month on the basis of new PPP levels. 10.3.1.6 PPP All three modelling approaches require additional inputs from the modellers to get results. The horizon (the time over which PPP is presumed to work) is one variable; perhaps more critical is the PPP reference levels. Calculating definitive PPP is not a science. While it is possible to analyse the price of traded goods, and determine exchange rates that equalise those prices, there are a wide variety of imponderables that make the calculation not definitive. Taxation

216

Currency Overlay

differences, product specification differences, differences in financial structure of the respective economies, transport and distribution hurdles, and so on.9 One way out of this uncertainty is perhaps to use the long-term inflation-adjusted historical average of exchange rates, on the basis that these will have oscillated round a long-term equilibrium (presumably something near to PPP). The problem with this is: what historical period do you use? If the period is contemporary with the testing period (in this case 1980–2002), then there is anachronistic data in the model. If it is earlier, it may be less relevant, and pre-1971, there were fixed exchange rates in which market clearing could not operate effectively. Box 10.6 Purchasing power parity PPP is an important theory for the determination of exchange rates. It is important because it is the bedrock of the argument as to whether exchange rates are over- or undervalued. The core theory runs as follows. A large part of industrialised economies’ GDP is tradeable goods (cars, food, drink, clothes, consumer durables). We can take a basket of these goods, and convert their domestic prices in different countries into a common currency (say the USD) at current exchange rates. This will show a dispersion of countries’ price levels – e.g. that a hamburger in Japan is twice as expensive as a hamburger is the US, implying that the yen is overvalued against the dollar. The PPP argument is that relative price levels will tend towards parity as arbitrage takes over. If an identical CD player can be sold for twice as much in Sweden as in Singapore, then traders will export more CD players to Sweden, and less to Singapore. If repeated across the economy, this will increase the balance of deficit in Sweden, and increase the balance of payment surplus in Singapore, depressing the Swedish exchange rate and raising the Singaporean, tending to equalise the prices. While superficially attractive, there are a number of major flaws in this theory. The key flaws from the economic perspectives are: non-tradeable country-specific factors like labour costs, taxes, property prices and infrastructure affect domestic prices, but are not arbitrageable. There are also the effects of tariffs, transport costs, differential product specification costs and differential indirect taxes on damping arbitrage for the elements which are tradeable. Finally, and perhaps most important, is the dominance of capital flows in the short-term determination of exchange rates. This can allow PPP overvaluation, together with large balance of payments deficits, to persist for many years without exchange rate movements to compensate (or even with perverse exchange rate movements). For a quantitative 100-year study, see Alan M. Taylor, A Century of Purchasing-Power Parity, National Bureau of Economic Research Working Paper No. w8012, November 2000. It also has a very good bibliography. I have chosen to test the following models on data 1980–2002, with PPP equilibrium levels calculated from 1971–1990 average actual price and exchange rate levels. This assumes that the secular differences in economic structure will be reflected in the 30-year averages. Hence if an exchange rate appears constantly overvalued in the OECD statistics (like, say, JPY), this systematic bias is eliminated by adjusting for the 30-year averages. In other words, the 30-year 9 The OECD (Organisation for Economic Co-operation & Development) publishes monthly PPP values for OECD member states. They publish, inter alia, the ratio of actual exchange rates to those that would give common price levels across member countries. Such published statistics, however, do not solve any of the problems listed above.

Active Currency Overlay – Management Styles

217

average PPP level for each currency is defined as 100. A currency is therefore overvalued if it is trading at an adjusted PPP level of >100, and undervalued if it is trading at a level of 10. Likewise, the trading models are all profitable, but with different horizons showing peak performance. It all looks good, but is it the makings of a successful fundamental model? Not necessarily. There are a number of points to make. The most obvious is that when the PPP equilibrium level is varied, to test the sensitivity of the model to this value (which should be important if the theory is right), two of the three trading models are insensitive to the changes. Both models still make money when every trade is ‘short’ of sterling, or alternatively every trade is ‘long’ of sterling. For this to be the case, the market must have exhibited a meanreverting element over this period (at least at the horizons tested). This means that the model ‘looks’ like a fundamental model, but is in fact a technical model, making money out of mean reversion. Table 10.1 Fundamental model results Type of model

Target to be modelled

External variables

Results

1 Correlation

Change in future currency surprise Change in future currency surprise

% spot over/undervalued vs. PPP % spot over/undervalued vs. PPP

Cumulative profits

% spot over/undervalued vs. PPP

>0.8 – peak horizon 85 months t-stat for PPP >10 depending on horizon Profits peak at 13 months horizon

Cumulative profits

% spot over/undervalued vs. PPP

Profits plateau between 15–30 months horizon

Cumulative profits

% spot over/undervalued vs. PPP

Profitable for all PPP levels

2 Regression 3 Trading model – fixed horizon forward position proportionate to ratio 4 Trading model – fixed horizon forward position +/− depending on over/undervaluation 5 Trading model – variable horizon forward position +/− depending on over/undervaluation

218

Currency Overlay

We can cast further doubt on the model by a split test – the model makes five times as much in 1980–91 as it does in 92–02, and in 98–02 makes almost nothing. Finally, this exercise has looked at one bilateral currency relationship. A fully-formed fundamental model needs to look at many more than one currency pair. It may need to rank the strength of the signals generated to ensure that risk is being most efficiently rewarded, and it will need to fit into the risk/return requirements that the investor demands. 10.3.1.8 Forward rate bias The foreign exchange market, like the commodities markets, not only has spot prices actively quoted and traded every day, but also a range of forward prices up to one year horizon. As we have seen, the forward discounts and premiums (vs. spot) are themselves a source of information, and since they are additional information outside the univariate price series, they probably have to be put in the ‘fundamental’ style pot. This is a little tenuous: most currency overlay managers would not call the forward rate bias a ‘fundamental’ effect, but in the arbitrary styles currently fashionable, FRB probably goes in this pot. I will not spell out here an active process which exploits the forward rate bias, since I covered this quite extensively in Chapter 9. 10.3.1.9 Summary on fundamental modelling I have not brought the modelling exercise to a conclusion. I have not looked at the volatilities of the value added, and so the quality of the value-added series is as yet untested. But the experience described above is common: powerful relationships at first look, and then with closer and closer attention the reasons for the relationships become less clear, or emerge as counterintuitive, or are ‘picking up’ the statistical influences of other variables. The evidence, which regrettably is mostly anecdotal, is that fundamental inputs into currency overlay are inadequate, on their own, to provide an effective active overlay process. The key shortcomings are instability of explanatory relationships, and the unresponsive nature of fundamental models. Instability is a very serious shortcoming indeed in modelling. It arises from explanatory models that are generally incomplete; that is, they are missing significant explanatory elements which, at least for some periods, are important contributors to overall price moves. Fundamental modelling faces the almost impossible task of including all these variables if a model is to be stationary (the opposite of unstable in this context). But as a modeller increases the number of these variables (e.g. is the Malaysian Central Bank active today Y/N?), the danger increases that the intuitive basis for the model – the core hypotheses – are diluted and lost. This will lead to high R 2 s and t-stats in the modelling period, and huge disappointment in the live period. Even if a robust stationary model can be constructed, there is the problem of ‘responsiveness’. The model constructed above had an element of responsiveness built in – the PPP over/undervaluation calculation is dependent on the spot rate, and this introduces an element of ‘lagged dependent variable’ into the equation. ‘Lagged dependent variable’ describes the situation where part of an explanatory variable (PPP over/undervaluation) is itself an earlier value (exchange rate) of the value we are trying to predict (also the exchange rate). Lagged dependent variables bring their own statistical problems, and we will look at these in Section 10.4, where lagged dependent variables are the sine qua non.

Active Currency Overlay – Management Styles

219

10.3.2 Technical Managers who use the ‘technical’ currency overlay style believe that the currency markets are not weak form efficient. In other words, they believe that the markets’ own price history has predictive power for future prices. I have already stated earlier at the start of this chapter that I believe the evidence strongly supports this position. The chapters (Chapter 4 and Chapter 9) that describe the structure of the currency market, and its dominance by non-profit sensitive participants, explain why this belief can be well founded, and that this inefficiency can persist. This forms the background to technical managers. The detail is largely statistical. If there is information in the price available at time t which is predictive (however indirectly) of future prices, then this is the basis for a successful technical model. We can explore a similar process to that undertaken by the fundamental modeller above. Let’s set the scene. 10.3.2.1 Hypothesis Let us suppose that the hypothesis is that exchange rates exhibit trends. This sounds like a rather simple statement, but establishing exactly what it means, and testing it, is far from simple. For example, the signals can be based on any price – spot is fine if it works, but the assessment of whether signals can make money must be based on currency surprise. Trends in one direction do not last for ever, and so there will be decisions to make about the horizon of any trend signal, and turning-point processes. 10.3.2.2 Inputs We need a daily database with spot and forward rates. We may need higher frequency data if intra-day strategies are to be modelled, and even higher frequency if any signal is to be ‘instantaneously’ actioned. We also need to pick a currency pair to start with. We will use again USD/GBP as an example, and data 1973–2002. 10.3.2.3 Correlation We can run correlations between the currency returns and prior period currency returns, or in algebraic form: Correlation (yt , yt−1 ), t = 1972–2002,

variable frequency

We can vary the horizons of the returns (i.e. one-day correlated with prior one-day, or five-day correlated with prior five-day) to see whether any autoregressive effect changes with horizon. Positive correlations imply trends (positive serial correlation), negative imply mean reversion (negative serial correlation). We are not constrained to the same horizons for the past returns and the future returns, i.e. we could test the correlation between one-day returns, and the prior three-day returns. We will explore this route of enquiry in more detail in Section 10.3.2.4. For this exercise, we can show the results from varying the horizon of the returns similarly for both series. If we use about 30 years of daily data, we have about 7550 points in the series. As the horizons extend beyond one day, the number of independent observations falls – two-day returns have 3775 independent observations, etc.

220

Currency Overlay

Correlation coefficients can be tested for statistical significance. The formula for this is: √  n−2 t =r √ 1 − r2 where t is a t-stat measure, r is the observed correlation coefficient and n is the number of pairs of observations. If t > 1.96, then the observed correlation is positive and significant at the 95% level (two-sided test). Similarly, if t < 1.96, then the correlation is significantly negative. For 7500 observations, the significance level is 0.022 (or −0.022) (i.e. apparently very low). As the horizon period rises, the level required for significance will also rise. We show the results for the 1973–2002 period in Figure 10.1. 0.16 0.14

95% Confidence limit

Correlation coefficient

0.12 0.10 0.08 0.06 0.04 0.02 0.00 0

5

10

15

20

25

30

35

40

Data frequency (days)

Figure 10.1 Correlation coefficient curve (USD/GBP, 1973–2002, lagged dependent variable, variable data frequency)

Only daily data appears to offer any real significance over this long data period, despite the marginal significance of 25-day data (or roughly monthly). All the horizons show positive correlation, indicating that this relationship has trended over history. However, if we were building a model to use in live trading, we would be building it over the ‘testing period’, and running it over a ‘live’ period, outside the testing period. This requires the relationship we establish in the testing period to persist into the live period. This ‘stability’ is core to the modelling process, and is discussed in Chapter 10 in more detail. For this example, we can look at the ‘split test’ of correlation: this is dividing the data into three roughly equal-length periods – 73–82, 83–92, 93–02. Each of these can be tested independently, as the graph in Figure 10.2 shows. Regrettably, this new information thoroughly undermines our previous hypothesis. It is not so much that the USD/GBP does not trend, because it clearly did in the 20 years 1973– 1992, but that it changed its behaviour in the last 10 years, and became a mean-reverting

Active Currency Overlay – Management Styles

221

0.3 73–82

Correlation coefficient

0.2

83–92

0.1

0.0

−0.1

93–02

−0.2 95% Confidence limits

− 0.3 0

5

10

15

20

25

30

35

40

Data frequency (days)

Figure 10.2 Correlation coefficient curves (USD/GBP, 1973–2002, lagged dependent variable, split tests, variable frequency)

currency with negative autoregressive behaviour (although at longer horizons than the trending behaviour). 10.3.2.4 Regression We can express a basic trends hypothesis in the following format: yt = f (yt−1 , yt−2 , yt−3 , . . . , yt−n , e) where yt = ct+n − ct , where ct = interest-rate-adjusted log currency index, and e is the error term (i.e. the unexplained variations in y). Expressed in words, this is a rather general model which says that returns in period t are (partially) explained by returns in periods t − 1, t − 2, t − 3, . . . , t − n. This type of model is called a lagged dependent variable model. The dependent variable is yt , called such because in the model context it is dependent on the explanatory variables. If we can get a dependent variable for a future period to be substantially explained by variables whose values we already know (lagged), then we have the makings of a forecasting model. We could also re-specify the model slightly into: yt = f (xt−1 , xt−2 , xt−3 , . . . , xt−n , e) where xt = pt+n − pt , and pt = log spot price. This might reflect more closely what drives the price – and despite appearances, this is still a lagged dependent variable model. It is usual to model y, the log currency surprise (sometimes called currency returns), rather than c, the log interest-rate-adjusted price. Returns are the rate of change of prices, and in continuous maths would be called the first differential. Modelling returns (rather than price)

222

Currency Overlay

is more useful because the random-walk-like nature of prices means that today’s price is always 95%+ determined by yesterday’s price, so price models give overwhelming (but rather indiscriminate) credence to lagged dependent variables. By contrast, even if there is serial correlation in currency returns, it is weak. So currency returns today are determined only to a small extent by yesterday’s returns: this is the value we want to get as high as possible. I have chosen to model log currency surprise (yt ) with five lagged dependent variables, each one period earlier than the other. I have used linear regression, the statistical details of which are available in any good econometric modelling textbook.10 The regression model is: yt = a + b1 yt−1 + b2 yt−2 + b3 yt−3 + b4 yt−4 b5 yt−5 + e where b1 , b2 , b3 , b4 and b5 are constants (parameters) and e is the error term (i.e. picking up all the unexplained variations. Again, like the correlation model, we can vary the frequency of the data – I have tested 1–30-day frequency. Just to be clear, a 30-day frequency model would have the dependent variable as the 30-day currency return (i.e. log currency surprise), let us say starting today and ending 30 days in the future. The first lagged dependent variable would be the 30-day currency return ending today, starting 30 days prior; the second lagged dependent variable would be the 30-day currency return ending 30 days prior, and starting 60 days prior, and so on. To complete the testing, I examined the regression results using only one lagged dependent variable – the lagged-one-period dependent variable. I then ranked the explanatory power of all the alternative regressions by an F-ratio – essentially a statistical test of how confident we can be that the amount of variation explained by the model is not zero. A high ratio means we can be very confident; confidence levels are given by the F-distribution. The model with the highest F-ratio, giving us the highest confidence that the model has genuine explanatory power, was the simplest – the single one-day lagged dependent variable on daily data. Summary results for this example regression model are shown in Table 10.2. To the uninitiated, it may be surprising that a very simple model gives us most confidence. It does not give us the best R 2 – a 5-lagged-dependent-variable 18-day frequency regression explains over five times as much of the variation. But as we add explanatory variables, we add significantly to the chance that their explanations are spurious. Table 10.2 Regression results (USD/GBP, 1973–2002, autoregressive model) Criteria for ‘best regression’ F-ratio F-test confidence Best frequency b1 coefficient value t-stat for b1 coefficient Best model specification

F-test 21.51 99.99964% 1 day +0.053(+ = trends; − = mean reverting) 4.4 Independent variables: single one-period lagged Zero intercept

Finally, we need to look at a split test. Having seen the results of the correlation split test, we can already guess the result: 1993–2002 will have a negative b1 coefficient. 10

For example Johnston and Dinardo, Econometric Methods, 4th edn, McGraw-Hill, 1996, ISBN 0-07-913121-2.

Active Currency Overlay – Management Styles

223

The results to prove it are shown in Table 10.3. We see that b1 is negative, but not significantly so over this period, and the F-ratio is not high enough for us to be at all confident that the model explains anything at all. Table 10.3 Regression results – split test (USD/GBP, 1993–2002, autoregressive model) Criteria for ‘best regression’ F-ratio F-test confidence b1 coefficient value t-stat for b1 coefficient

F-test 1.70 81% −0.028 (+ = trends; − = mean reverting) −1.42

Regression models are explanatory, not trading processes. To convert significant regression results into a trading model, the modeller is required to develop a process. This might be a ‘momentum’ model, in which the parameters that are identified in a regression as having significant explanatory power are then pressed into service to take an active currency position (proportionately ‘long’ for predicted price rises and proportionately ‘short’ for predicted price falls). I show results of a simple version of such a model below. 10.3.2.5 Trading models The rules of trading models can be infinitely varied. Their purpose is to codify actions that a currency manager needs to take at time t to ‘maximise’ the chance that the result, once the position is closed, is a profit. I have constructed and tested two models. The first is a moving average model, the second is a momentum model. Moving average model: A moving average model is one where a simple trading rule is established: buy the currency that is rising at the moment a shorter trailing moving average of the spot rate crosses a longer trailing moving average. Such a model will exploit positive serial correlation if it exists; the reverse rule would be applied if it is believed that markets are meanreverting. Note that it is perfectly possible for markets to be both trending and mean-reverting at the same time. The effects can apply at different horizons – say trending up to 3 months, but mean-reverting at 2 years. I will be testing shorter effects in this modelling. Moving averages are expressed in days, and in modelling this kind of process, one can visualise the shorter trailing moving average (say 2 days) being a damped proxy for the current price, and the longer trailing moving average (say 18 days) being a proxy for a ‘trend’ line. When the shorter crosses the longer, the model is implicitly saying that the trend has changed, and therefore the position should too. I have constructed a moving average model which uses the spot rate as the trend indicator (i.e. as the constituent parts of the moving averages), and which takes a 1 or −1 position (long and short) depending on the relative values of the moving averages. However, the profits or losses from this position are recorded from the subsequent currency surprise, not the spot rate. It is in theory possible (although only with wide interest rate differentials) that the moving average model could correctly predict the spot rate, but that the position would be loss-making because of the interest rate differential erosion. The model also builds in transaction costs

224

Currency Overlay

Box 10.7 Mean reversion This is a common expression used to describe a particular type of market behaviour, but what exactly does it mean? It is a rather loose expression, but in ‘non-mathematical’ speak it means that (a) market movements are not random and (b) the market is more likely to reverse a previous move than continue it. The horizon over which mean reversion applies is very important in determining its effect on market prices, and the kind of active process to exploit it. Very short horizon mean reversion This is where the mean reversion applies over the shortest of horizons – say one day. This means that if the market goes up one day, it is more likely to reverse that move tomorrow rather than continue it. This behaviour is characteristic of illiquid markets, and the ‘choppy’ market prices this brings about are often the result of the absorption process of large orders in the market. Mid-horizon (moving average) reversion Under this model, the market has a ‘memory’ of a moving average of past prices, and there is an ‘elastic band’ which is trying to move the market level back towards these historical levels. The moving average can be of a wide range of maturities. It is supportable as a proposition that exchange rates are mean-reverting over 5-year horizons. Fixed mean Under this model, the mean is not a moving average of historic prices, but a broadly fixed value. The most obvious potential example of this is real (inflation-adjusted) interest rates; they vary day-by-day and year-by-year, but they tend to revert to a value centred around (say) 3%. This is perhaps the purest example of mean reversion. Mean reversion and trending can co-exist. Indeed there is strong evidence that exchange rates are trending over ‘market memory’ horizons – say 0–1 years, but mean-reverting over 5-year horizons. This is not particularly surprising: market movements are the result of the collective behaviour of the market’s participants, and the complex interactions of sentiment and underlying economic variables may perfectly well have different horizons. for changes in the position. This kind of detail is important in modelling, since one key and necessary condition of a model that proves successful in practice is one that fully represents all the factors encountered in the live trading process. The results for this model are shown in Table 10.4. I have used ‘information ratios’ (broadly speaking, annualised return divided by annualised standard deviation of returns – see Box 10.8) as the ‘performance’ result, and this is calculated from the daily value-added calculated from this model. From the table, the reader can see that the best result was from the 2- and 18-day combination. The table also neatly illustrates the effect of transaction costs. The values for 10- and 12-day moving averages are repeated in both row and column. The difference of 0.2 (i.e. 0.52 − 0.72) information ratio is twice the effect of the transaction costs, which is therefore 0.1 information

Active Currency Overlay – Management Styles

225

Table 10.4 Information ratios of moving average model, daily data (1973–2002) Longer moving average (days) Shorter moving average (days)

10

1 2 4 6 8 10 12

0.58 0.56 0.61 0.48 0.45

Box 10.8

−0.72

12

14

16

18

20

22

0.54 0.60 0.65 0.60 0.59 0.52

0.64 0.63 0.58 0.52 0.42 0.44 0.23

0.73 0.75 0.60 0.42 0.35 0.38 0.33

0.71 0.76 0.43 0.29 0.27 0.32 0.37

0.67 0.61 0.41 0.36 0.31 0.40 0.28

0.66 0.52 0.39 0.34 0.30 0.33 0.18

Information ratio

The expression ‘information ratio’ was first coined by Nobel Laureate William Sharpe as a useful way of measuring the ‘skill’ of active managers. The principle on which it rests, ‘risk-adjusted excess return’, is the same as his parallel measure for securities – the Sharpe ratio. The information ratio is the average added value of a portfolio (portfolio return less benchmark return) divided by the tracking error (see Box 6.1). This value (generally lying in the −1 to +1 area) records how much risk a manager is taking to achieve the recorded value added. Information ratios are calculated on an annual basis, so if quarterly, monthly or higher frequency data is used, it must be annualised first. Information ratios greater than 0 indicate that a manager might have demonstrated skill. But active managers have to deviate from benchmark returns to add any value at all, and so there is, on the face of it, a roughly 50% chance that a manager with an information ratio greater than 0 will have achieved it by chance. We can test this by using the property of the very useful Central Limit Theorem. In summary, if we multiply the information ratio by the square root of the number of years for which the averages are calculated, we will get a ‘t-stat’. Roughly speaking, if this number is greater than 1.7, it tells us that there is a 95% chance that the manager has demonstrated skill not luck. There is a ‘rule of thumb’ in evaluating managers that an information ratio of 0.5 is good, and 0.75 is excellent. It is highly unlikely that any active manager can demonstrate a prolonged information ratio of 1. An active manager with a (more typical) 0.4 information ratio would have to show a 16-year track record for this to prove skill, whereas a 0.75 information ratio manager could prove this in 5 years. ratio. The effect of transaction costs is less with wider differences of moving average horizon – at 2 and 18 it is about 0.06 information ratio. This apparently stable state of affairs is, unfortunately, upset when we apply the same split test as in the case of correlation above. If we take the best moving averages (2 and 18), we can calculate the information ratios for each of the split periods (Table 10.5). So the last 10 years has shown negative returns for this strategy, while the previous 20 years gave a staggering

226

Currency Overlay Table 10.5 Moving average model – USD/GBP, 2- and 18-day, information ratios for different periods Period Whole period (1973–2002) 1973–82 1983–92 1993–2002

Information ratio 0.76 1.52 1.22 −0.54

average of an information ratio of 1.37. This ‘change of market nature’ fits in with the negative serial correlation we detected. The temptation is to attempt to develop the next level of refinement to provide a filter to give ‘early warning’ of the change from positive to negative serial correlation. This kind of thinking is appealing, but has to be treated with caution. This is for the reasons set out earlier in the chapter. It encourages the idea that with enough explanatory variables, all the ‘moving parts’ of this series’ generator will be captured. A better response, in my opinion, is for the modeller to take the trouble to try to understand, if such a thing is possible, the reason(s) for positive serial correlation appearing in the data in the first place. If these are accurately identified, then if these factors change or disappear, independently of the data, the modeller can progress to predicting such market behaviour changes with a well-founded likelihood of success. Momentum model: A momentum model can take many forms. The moving average model is a particular type of momentum model. In this section, we will look at another type of model, which in this example is our regression model above. I will illustrate one class of models in the large universe of possible autoregressive models. Interested readers can find a great deal on univariate11 time series analysis, as it is called, in econometric and statistics textbooks. It is worth noting that this universe also covers autoregressive second and higher moments – like autoregressive volatility12 and so on. Finally, staying with univariate modelling, we could (although not here) move away from the strictly statistical into pattern recognition and pattern matching. This moves us into the world of charts and chartists, and more esoteric processes like fractals.13 In an effort to keep the example manageably simple, for these illustrative purposes I have built a momentum trading model that applies the ‘best model’ from the regression analysis above, and turns it into a signal-generation process for position-taking. This is not the principal technique applied to building momentum models, which will be simulation techniques (see Box 10.9). However, in this example, I have chosen to set the size of the position (long or short of GBP) to the size of the predicted move from the autoregressive model. Since the ‘best’ model is so simple, this is actually proportionate to the size of the move the previous day. The model is reset daily, and gives daily returns, which makes it easy to calculate information ratios. I have chosen to show the results with and without transaction costs. This is because a very ‘hair trigger’ model like this, which changes the position every day, and which tries to exploit a very 11

Univariate means that there is one time series (the price), and all modelling is conducted on the basis of this information alone. For example GARCH (Generalised Auto-Regressive Conditional Heteroscedasticity), a univariate modelling process for nonstationary volatility. 13 Fractals are self-similar geometric shapes. ‘Self-similar’ means that they look the same however much you enlarge or reduce them. There is a lot on fractals in maths and art textbooks, and on the web. 12

Active Currency Overlay – Management Styles

Box 10.9

227

Simulations

A significant part of the research and development process of active currency overlay managers is spent building and testing trading models. These are routinely tested in historical simulations. In a simulation, a model which has been developed using a range of historical data (usually up to the present time) is ‘run’ over history to simulate the trades it would have generated had it been live at the time. These simulations can be used to refine models to produce very high value added and low tracking errors (leading to high information ratios) for the whole period tested. Well-designed simulations are a very valuable tool for managers in their active process development efforts. However, good simulation design requires:

r Full accounting for transactions costs; r Care over the exact historical timing of information arrival (to avoid trades being ‘executed’ prior to the arrival of the relevant information);

r High-quality price information; r Care taken to ensure that the simulated trading process is replicable in live trading (i.e. frequency of price monitoring, lags in executing real transactions). However, the most important aspect of simulations is that they will generally produce historical results that are consistently better than those achievable in live trading. Why is this? In the currency markets, it is not primarily due to lack of liquidity or optimistic pricing in the model. Rather, it is the conceptual problem of model selection. Models which survive the ‘cuts’ imposed by the testing and simulation process will be ‘better’ (i.e. have better historical returns) than those that do not. Surviving models may have a proportion of their good performance explained by real and stable inefficiencies, but also a proportion explained by chance. The random element is likely to revert to the mean (i.e. zero performance) for the future (live) period; only the ‘real’ element will contribute any value added. The disappointment engendered by underperformance of live models has tarnished simulations as a tool. This is shooting the messenger – the problem is not the simulations but the modelling process. The most effective modellers are those that understand the survival problem, and build models which may on the face of it produce inferior historical returns, but which continue to produce similar returns in live trading. weak effect over very short intervals of time (one day), has a very high cost/benefit ratio. This is clearly shown in the results (Table 10.6). The results are not particularly appealing – transactions costs (even at the low level achievable in the currency markets, I have assumed 3 bp round-turn on outright forwards) have a very debilitating effect on results. The 93–02 period stands out as being pretty disastrous. Clearly there is some work to do on this model to make it remotely usable. 10.3.2.6 Summary technical modelling results Summary results are provided in Table 10.7. As we have seen, behind these superficially attractive results is a great deal of non-stationarity. 30 years of daily data is a huge amount

228

Currency Overlay Table 10.6 Momentum model – USD/GBP, ‘best fit’ regression, information ratios for different periods Period

Information ratio (without transactions costs)

Information ratio (with transactions costs)

0.67 0.50 1.37 −0.36

0.36 0.21 1.08 −0.76

Whole period (1973–2002) 1973–82 1983–92 1993–2002

Table 10.7 Technical model results Type of model 1 Correlation 2 Regression 3 Trading model – moving averages 4 Trading model – momentum

Target to be modelled

External variables

Results

Change in future currency surprise Change in future currency surprise Cumulative profits

Lagged dependent variable Lagged dependent variable(s) Lagged dependent variable Lagged dependent variable

t-stat = 4.65 at 1-day autoregressive Best F-ratio = 21.5 at 1-day lag IR peaks at 0.76 at 2- & 18-day MAs IR peaks at 0.36 at 1-day frequency

Cumulative profits

(c. 7550 days), and the models have no difficulty whatever in finding highly significant explanatory models in all of the techniques. However, none of these techniques, as they stand, would have allowed their users to make money in the last 10 years – in fact they would have lost money. In this exercise, I have looked at one exchange rate series over a particular period of time. Of the major currency pairs, USD/GBP happens to exhibit the most extreme changes to the nature of its distribution. Other pairs have changed less, or not at all, or have become more trending with the passage of time. What this exercise will have illustrated, I hope, is that technical modelling can demonstrate persistent inefficiencies, but that exploiting these inefficiencies requires a well-founded belief in the underlying causes of the inefficiency as well as technical modelling expertise. So successful modellers under this reasoning should be experts in the foreign exchange markets first, and mathematicians second, not the other way round. I have concentrated on one currency pair, but of course there are many. One of the advantages of technical modelling is that it can choose, on the basis of recent currency behaviour, which pairs are conforming to the core model. This is a further powerful tool, which, in the right hands, can make the technical style one of the most powerful. Finally, technical model building requires an understanding of the needs and constraints of the client. Unconstrained technical models can undoubtedly make money over time, but if they require a US-based client to take positions, long or short, in, say, JPY/GBP, is this acceptable? Certainly a conventional overlay mandate, as we have seen above, imposes certain constraints on the manager: 0–100% hedge ratios, all positions being part of a pair with the base currency, and the size of the position being determined by the underlying asset holdings in those currencies.

Active Currency Overlay – Management Styles

229

10.3.3 Option-based The currency option market developed in the mid-1980s, and is now large and very liquid. It certainly qualifies on these grounds for inclusion as a suitable instrument for the currency overlay manager. Options are instruments which, as we have seen in Chapter 3, give the option holder rights but no obligations in return for the up-front payment of a fee (the premium). Conversely, options confer on the writer (seller) obligations but no rights in return for receipt of the fee. Given the complexity of the concepts, and the maths surrounding options, the reader should have familiarised himself with the contents of Chapter 3 before embarking on this section. This section looks at the additional dimensions of decision-making that options give the overlay manager, and the routes that might allow systematic exploitation for value added. Underlying these descriptions is the presumption, given very strong credence in the earlier sections of this chapter and in Chapter 9, that currency markets are not efficient, and therefore that results from analysis and active management can be positive on average over time. 10.3.3.1 Passive option buying in serially correlated markets Passive option buying is not a popular currency management style. It looks too ‘passive’ for most active currency managers, and perhaps, more pertinently, their clients. However, there is a perfectly respectable and mathematically sound case to be made that passive option buying has positive expected value added. We have established from a variety of evidence offered in previous chapters that many currency pairs exhibit trends. I have argued that this trending behaviour has its foundations in the structure of the foreign exchange market. It has shown persistence, and the structure of the foreign exchange market has not materially changed over the period – it is still dominated by profit-insensitive players. Let us take an example of how such an option-buying process might work. We will make the working assumption that implied option volatilities are equal historical volatilities calculated from the daily vols, and we will test this assumption later. We will look at 12-month options, since the evidence of trending indicates that we should extend the option maturities as far as possible. What are we trying to exploit? Figure 10.3, reproduced from Figure 9.1, shows the volatility differentials for the major currency pairs. We will be trying to establish whether we can buy options at the daily volatility generated Black–Scholes model premium, and hold them to receive an average payoff that reflects the annual volatility. Let’s start with the theory, which I will summarise in a mathematically simplified way. The Black–Scholes model formula for an at-the-money forward option (put or call) boils down to: FV 1-year option premiums = (0.796/2) × volatility

(10.1)

Taking each of the terms in turn: FV (future value) option premium means the amount payable contemporarily with the payoff value, i.e. at maturity and not up front. The up-front premium payable would be the FV divided by the 1-year interest rate, but then to compare with the payoff value we would have to multiply by the interest rate (because the option writer holds the money for one year before it is needed).

230

Currency Overlay 18%

Annualised daily

16%

Annual

Volatility % p. a.

14% 12% 10% 8% 6% 4% 2% 0% JPY/ USD

JPY/ DEM

JPY/ GBP

DEM/ USD

DEM/ GBP

USD/ GBP

Figure 10.3 Volatility measured at different horizons – volatility of currency surprise (1981–2002), daily volatility annualised and annual volatility

0.796 is the multiple applied to volatility (standard deviation) to calculate the absolute average outcome of a normal distribution (see footnote 2 in Chapter 6); /2 because half the time a call/put expires out-of-the-money and gives zero payoff value. Volatility is calculated from the log of currency returns (i.e. currency surprise), annualised √ by calculating daily volatility and multiplying by 262. I will convert log values back to percentages after all the algebraic manipulation required. Using exactly the same logic, and formula, we can also calculate the average expected payoff value. This will have all the same elements of the previous formula, except that the volatility is the average 1-year volatility. Average historical outcome: If we take the history of DEM/EUR against the USD (I will call it DEM/USD here) as an example, Figure 10.3 tells us the daily annualised percentage volatility calculated from the logs is 11.69%. The average option premium under our assumptions is therefore: exp[(0.796/2) × log(1 + 11.69%)] − 1 = 4.50% The average annual volatility for DEM/USD (also from Figure 10.3) is 14.38%. It follows that the calculated average payoff value is: exp[(0.796/2) × log(1 + 14.38%)] − 1 = 5.49% So on the face of it, the trending behaviour observed may be a cause of ‘undervaluation’ of option premiums, and therefore render passive option buying a profitable strategy. Implied volatilities – the real test: I have collected a nearly 6-year series of daily implied volatility quotes from a major currency option provider bank for DEM/USD. An example of these quotes is shown in Figure 3.5. However, we need to test 12m option quotes (not 3m option quotes as shown in the graph). I will use the 12m offer quotes, the implied volatility the bank is using to price its options offered to customers. Its bid volatilities average 0.25% lower than the offer volatilities over this period. Its volatilities are expressed as percentages,

Active Currency Overlay – Management Styles

231

although as explained in Appendix 2, these are really logs (i.e. percentages in continuous time). Banks do not exponentiate their option calculations based on logs back into proper percentages. However, since they charge clients option premiums based on the quoted percentage multiplied by the option principal, I will treat the quotes as percentages. Applying the test: We are going to calculate option premiums, and the actual payoff values experienced, for DEM/USD 1997–2002 daily: 1. We can calculate the FVs of at-the-money option premiums from the implied vols without interest rates using equation (10.1). Puts and calls will be the same price because of arbitrage (put + forward purchase = call, although this arbitrage only works at-the-money forward). This will give us a series of option premiums. 2. We can double this cost, because in the ‘value for money’ test we are going to buy doubles (i.e. buying both a put and a call). This will take away any dependency on arbitrary direction of exchange rate moves from the assessment. 3. Then we need to calculate the payoff values for one-year DEM/USD puts and calls. Note that I have not distinguished the numerator and the denominator, and as long as we work in ‘log space’, we do not have to – the results are the same. 4. Payoff values are not guesswork or averages. They are fact. Here is an example. One-year DEM/USD LHS forward rate on 2 Jan 1997 was 1.5009. This was the price to buy DEM, sell USD. One year later, on 2 Jan 1998, the LHS spot was 1.8042. The payoff in log terms was log(1.8042/1.5009) = 0.1841 for the profitable option (in this case the DEM put/dollar call), and zero for the DEM call/Dollar put. I take 4 bp off the profitable log value to account for the fact that the put and call strikes will be marginally different, reflecting a 4 bp one-year outright bid/offer spread, and the cost of closing the profitable option at spot. This leaves us 0.1837 in log terms as the double profit. 5. Converted back to percentages, the payoff was exp(0.1841) − 1 = 20.16%. 6. The premium for a single option from the volatility quote in this case was 3.52%, or 7.04% for the double. The profit on this observation (i.e. a double bought on 2 Jan 1997) was 20.16% − 7.04% = 13.12%, or 6.56% for the single equivalent. Figure 10.4 shows the net value of the single option equivalent payoff values minus option premiums for the purchase periods Jan 97–Nov 2001 (maturing Nov 2002). We can clearly see that this net payoff is quite volatile, but that it delivers +44 bp of value added over the period. This is not a spurious finding – it fits in with the trending nature of data we have studied, and what we know about how banks price option premiums. Table 10.8 gives summary information. Table 10.8 One-year DEM/USD options, averages 1997–2001 (purchase dates) Average option premium (FV) Average (put + call)/2 payoff value Average gain per option % Max gain % Max loss %

4.38% 4.82% 0.44% 11.93% −5.61%

This result arises from one definitive value (payoff) and one less transparent value (option premium). How can we be more confident about how option premiums are quoted? I have compared the option premiums (from implied vols quoted by a major bank) with historical volatilities calculated from spot rates on a rolling backward-looking 6-month period. So for

232

Currency Overlay 14.0% 12.0% 10.0%

Average = +0.44% 8.0%

Net profit %

6.0% 4.0% 2.0% 0.0% −2.0% −4.0% −6.0% −8.0%

1997

1998

1999

2000

2001

Figure 10.4 One-year DEM/USD put + call options – payoffs less premiums ($/DEM (latterly EUR) options, 97–02, 12m maturity)

2 Jan 1997, the log of the spot changes 2 Jul 1996–2 Jan 1997 would be the source data, and this sample’s standard deviation would be annualised, and used as the source to calculate the ‘historical’ option premium using equation (10.1). The graph of the implied (quoted) and the historical rates is shown in Figure 10.5. Quoted premiums are clearly closely based on historical volatilities, over this period they have averaged 4.38% vs. 4.21% for the historically calculated – a difference of 17 bp. All of the above is an example; it covers one particular period, for one currency pair, neither the most trending nor the least on the basis of Figure 10.3. It does illustrate, however, that there is a potentially interesting opportunity for overlay managers in this area of ‘passive’ option buying. 10.3.3.2 ‘Active’ option trading The Black–Scholes formulation assumes constant volatility. If this assumption is relaxed, as it plainly is in real markets, then there is a premium placed on successful prediction of volatility. 10.3.3.3 Volatility forecasting Volatility, more specifically implied volatility, has become a recognised market in the world of option trading. Option traders, when asked what they do, will often reply that they ‘trade volatility’. Indeed several banks offer OTC products which allow their customers to buy and sell ‘volatility’ directly. There are several issues that surround volatility trading. The first is that what is traded is implied volatility, not historical volatility. In a sense this makes the market peculiarly circular. Implied volatility is, in effect, a forecast of future volatility. Any manager wishing to successfully ‘trade’ volatility is being required to forecast future changes in the forecasts of

Active Currency Overlay – Management Styles

233

% Premium

6.50%

6.00%

Quoted premiums

5.50%

Calculated historical premiums

5.00%

4.50%

4.00%

3.50%

3.00%

2.50% Jan 97

Jan 98

Jan 99

Jan 00

Jan 01

Jan 02

Figure 10.5 Quoted and historical volatility option premiums (DEM (EUR)/USD, 1997–2002, 1-year options, ATMF, FV premiums)

volatility. Actual volatility may not come into the equation at all! The second point is that volatility is an artificial construct, predicated on (almost certainly violated) assumptions. ‘Volatility’ is traded because the Black–Scholes formulation assumes a lognormal random walk for the underlying markets, and volatility is the annualised standard deviation of these changes. However, as we have seen above, violation of the Black–Scholes assumptions (in particular uncorrelated random changes) can radically affect even the straightforward calculation of volatility. Turning to implied volatility forecasting, there is very little academic literature on this area, particularly in foreign exchange. There have been no systematic studies of implied volatility forecasting reliability or success, and the vast majority of the (mainly bank-sponsored) research in this area has concentrated on the relationship between historic volatility and implied volatility. But for forecasting purposes, this is of marginal use, since even a reliable implied volatility model based on historic volatility simply shifts the burden of forecasting to another variable – historic volatility. Historic volatility (a misnomer since the history may be only a few seconds old) is an average of indeterminate length of the magnitude of changes in the underlying market. Historic volatility can be calculated as the observed standard deviation of a sample of changes in recent history, and this is in practice applied with the assumption that the true (underlying) standard deviation has been constant throughout the sampling period. With increasing length of history, this would be fine if volatility were indeed constant (in line with Black–Scholes assumptions), but it is not fine if population volatility is either non-normal, randomly variable over time, or predictably variable based on other inputs. Over the past few years modellers have recognised that the underlying variance/volatility is not constant and have contrived to use more sophisticated models to deal with this. GARCH models forecast underlying variance as a (complex) function of past values. This is a more

234

Currency Overlay

realistic standpoint but unfortunately leads to an exponential growth in the number of parameters which need to be estimated (are they variable too?, when do we stop?), with all the attendant drawbacks. 10.3.3.4 Other active option strategies Putting aside volatility forecasting, what else can an active manager do with options to enhance returns? Banks will trade options profitably by making prices, not taking them. A currency manager, without a customer-orientated dealing room, client base, brand and market position and capital, cannot do this. A manager may use options as a downside-controlled way to take currency market bets in either direction, but this is not using option expertise so much as general currency market expertise to add value. 10.3.4 Dynamic 10.3.4.1 Option replication Very early in the history of currency risk management, in the early 1980s, before currency options became available, currency managers recognised that option-like payoffs were an attractive goal. We have seen in the earlier sections on option models that Black–Scholes’ 1973 paper asserted that it was possible to exactly replicate an option position with a dynamically adjusted portfolio of the underlying securities (or forward contracts in this case). Various assumptions violations mean that this assertion is not, in practice, true. Nevertheless, some sort of approximation is possible, and this section explores the various approaches managers have taken to this approximation. 10.3.4.2 Dynamic hedging overlay style As the OTC option market in currency options started in the second half of 1984, so the moneycentre banks who were offering them began to more or less crudely hedge the resulting exposure using dynamic, or ‘delta’, hedging techniques. The concept of delta hedging is broadly speaking to hold a position in the underlying market whose change in marked-to-market valuation per small (1 bp) movement in the underlying market is equal and opposite to the change in markedto-market valuation of the option sold. A bank that sells a large portfolio of options in one currency pair can combine these to create a portfolio delta, which is the elasticity of the whole portfolio of options with respect to a 1 bp movement in the underlying market. As an example, an outstanding option sold to a third party, say a $1m call against a third currency, could have a marked-to-market valuation which varies by $350 for a 1 bp movement in the exchange rate. Since a $1m position in the underlying market would have a $100 change in marked-to-market valuation per 1 bp move, this would be called a delta of 0.35 or 35%. Deltas will range between zero and one for ‘vanilla’ options, but can vary widely and wildly for exotic options. There is a common further measure of the shape of the delta curve with respect to the exchange rate which measures its instantaneous ‘curvature’, and this is called the gamma. There are a whole lot more Greek letters denoting the elasticities of the option valuation with respect to time, volatility, etc., but they are of interest more to bank risk managers than to the general currency overlay reader.

Active Currency Overlay – Management Styles

235

The argument for currency overlay managers offering a dynamic hedging style of currency risk management runs like this. Buying currency options to protect against adverse currency movements in the fund context is attractive for two reasons: it reduces volatility if correctly targeted and benchmarked, and it satisfies the asymmetric risk-aversion utility curve that meanvariance analysis is poor at coping with. From the evidence presented above, it can also exploit trending markets to add value. But a manager has to buy currency options from banks. This is true whether or not they are OTC or exchange-traded, since the market-makers on the major exchanges are the major money-centre banks. Banks sell currency options with the prospect of securing a profit on the combination of the premium received for the sale and their cost of covering their risk. They can cover their risk in one of two ways: buying options from third parties or running a delta hedge in the underlying market. Since there are no natural writers of currency options who write them without a delta hedge (in contrast to equity options where institutional holders may well write ‘covered’ options), banks generally are the ultimate writers of currency options. One way or another, therefore, options written to customers will be matched by a delta hedge run by a bank with the prospect of profit to the bank. Delta-hedging currency managers (or dynamic hedgers) offer, at the least, a delta hedge run on behalf of the investor to reduce the cost of the option premium by the amount of the prospective bank profit. They may also offer other advantages: flexibility in tailoring specific option specifications; high liquidity of the forward market compared to the option market; cash flow improvement (no up-front option premiums); and less sensitivity to market volatility in marked-to-market valuation. In a perfect Black–Scholes world these advantages would be unsullied by any disadvantages. The Black–Scholes model tells us that any option can be exactly and risklessly replicated by a portfolio of the underlying security (foreign currency or forward contracts in this case). As we have seen earlier in this chapter, however, Black–Scholes markets do not exist, and one of the effects of this is that option replication is not a riskless exercise. What is the effect of this in practice? To simplify the discussion, I propose to discuss the replication of one vanilla option with the following specification: Put currency Call currency Put amount Strike rate Initiation date Expiry date Maturity date European style (i.e. exercisable only at expiry) Premium (% dollar amount) Premium payment date

Japanese yen US dollar 1bn yen 100 yen/$ 1 Jan 1996 1 Jul 1996 (date for exercise) 3 Jul 1996 (date for payoff cash flows)

3.02% (=$302 000) 3 Jan 1996

Suppose that an investment fund wishes to ‘buy’ this option on 1 Jan 1996. If it wishes to buy it by replication and not in the OTC market, what will a typical dynamic hedging currency overlay manager do? At the moment when the ‘option’ is established, the overlay manager will use a Black–Scholes model variant (a variant to cope with forward contracts with no yield and no cash requirement), input the option specification and current market data, particularly the 6-months forward yen/$ rate and a volatility value. The volatility value may be current

236

Currency Overlay

implied volatilities, historical volatilities or some model value from the manager’s volatility forecasting model. The model will provide a series of outputs, the most important of which is the sensitivity of the marked-to-market valuation (in effect the option premium) to changes in the yen/$ exchange rate – the hedge ratio or delta. If this comes out at, say, 45.6% on the first data entry, then the overlay manager will sell 45.6% × yen 1bn = yen 456m against dollars in the forward market. The manager may choose to sell at any forward date, but the obvious ‘default’ value would be to sell for a maturity of 3 Jul 1996. What happens next is very much manager-specific, but Black–Scholes theory requires that the manager then conducts continuous monitoring of the exchange rate, and continually adjusts the hedge ratio to stay in line with model output, which will vary as the exchange rate and volatility changes and time passes. Black–Scholes require that this process is literally continuous, so that thousands of deals could in theory be conducted in the space of a second, or indeed in a nano-second. This requirement is clearly impractical, so the manager must embark on a series of approximations to get as close to this as possible. Firstly, a manager must decide on monitoring frequency and procedures. Does monitoring take place at fixed frequency: once a second, once a minute, once an hour, once a day, once a week? If monitoring is fixed frequency, what about 24-hour dealing? What about illiquid times of day, when perhaps only a thin Middle Eastern market is available? What about weekends, what about Bank Holidays? If monitoring is not planned to be fixed frequency, how is it conducted? If it is triggered by a certain level of exchange rate movement, how is this movement monitored, continuously? If monitoring is high frequency, does this always trigger adjustments in the hedge position, even if these are small? Are there buffer exchange rate zones or buffer dealing sizes that keep dealing frequency under control? How is volatility monitored? Do market implied volatilities drive the hedge ratio, or historical or model output volatilities? The variety of answers to these and more questions provide ample material for manager differentiation – and many of the issues here have been dealt with in other contexts earlier in the book. But there is a straightforward test that a customer can apply. Has the option replicator succeeded in replicating the payoff value of the option being replicated, and at what cost? The arithmetic of this is simple. Take the two possible states that an option can expire in: in-the-money and out-of-the-money. If the option expires out-of-the-money, then the payoff value is zero. In this case the cost of the dynamic hedge is the closed sum of the dollar values of all the forward contracts relating to the option. If the option expires in-the-money, then the payoff value will be positive. If the close-out spot rate on 1 Jul 1996 is 108, then the payoff is $740 740.74 [=(1bn/100) − (1bn/108)]. The cost of the dynamic hedge is then the difference between the closed sum of all the associated forward deals, and $740 740.74. As an example, if the delta hedge yields $550 000, then the cost of replication was $190 740.74 (=$740 740.74 − $550 000) or 1.91% of the dollar principal. The key question that arises from the process described is whether it consistently reduces that cost below buying the equivalent options in the market, and how wide is the variability of the cost. The evidence is difficult to find, in particular since option replication either by banks or by overlay managers is a commercial undertaking not open to public scrutiny. Under some fairly rigid assumptions about the monitoring process, however, it is possible to simulate the results of delta hedging using actual option premiums and underlying market rate histories. It appears from this slim evidence that delta hedging under these rigid assumptions is not materially lower cost than typical option premium offer prices, but nor does it have materially higher cost variance. Any cost savings experienced by dynamic overlay customers are likely to be the result of the managers’ proprietary refinements.

Active Currency Overlay – Management Styles

237

Dynamic hedging can also be operated entirely outside the Black–Scholes model environment, and a subclass of dynamic overlay managers do this. Under their processes there is no assumption of lognormality, of continuous markets, or even of volatility levels. Their payoff outcomes are identical to delta hedgers, but the composition of their costs is different. Their costs will be higher or lower than delta hedgers, dependent on whether their models more or less effectively exploit market non-randomness. (Note that with pure lognormal random walk markets, all strategies are ultimately identical at zero net return less transaction costs.) Dynamic currency overlay managers (or option replicators) represent one of the largest style classes of currency overlay managers. Many overlay managers will not use this style exclusively, but many rely on it as the core of their active process.

11 Active Currency Overlay – Evidence of Performance This is a short chapter. There is now some evidence of the performance of currency managers, but much of it is proprietary, and it ages quickly. In summary, however, there is good evidence that the universe of currency managers have on average delivered outperformance against passive benchmarks over the 14-year period 1989–2002. There seems to be some ‘reversion towards the mean’, in that earlier studies showed higher outperformance than later studies, and as the universe has expanded.

11.1 SURVEYS 11.1.1 Currency overlay performance surveys Currency overlay as a recognised activity in the institutional investment market started in 1985, when this firm (Record Currency Management) secured a mandate to actively hedge the US dollar exposure of the Water Authorities Superannuation Fund – then a public UK fund. Several more mandates followed in the UK in that and subsequent years. The earliest US mandates were 1988/9. The sector grew quite rapidly, largely on mandates from the larger US public and private sector funds. It began to spread to continental Europe, Australia and the Far East in the mid-1990s. The UK is still relatively undeveloped in currency overlay, despite having the second largest pool of defined benefit pension assets after the US. Performance measurement of currency overlay managers is a relatively recent phenomenon, but there are now at least three sources from which we can draw data. 11.1.1.1 The ‘Strange Report’ In 1997, an ex-currency overlay manager, Brian Strange, then working in his own small consultancy, Currency Performance Analytics, persuaded 11 currency overlay firms to submit detailed client-by-client data on their performance (vs. benchmark) for all their currency overlay accounts, live and terminated. A further three submitted composite data. The subsequent report, published May 1998, showed that, on average, currency overlay managers had added about 1.9% p.a. in the period since 1989, the earliest data submitted. The average tracking error in this study was 3.5%, giving an average information ratio of 0.54. The total number of currency overlay accounts was 152, and from this client base 1783 client-quarters were analysed. A number of different ways of segmenting the data were illustrated, and in each a story of consistent manager behaviour and consistent value-added emerged. There was little in-depth analysis of why a universe of specialist managers could perform so well en masse, but it was noted that a significant proportion of the client-quarters covered the 3-year 1995–97 period, when the dollar was broadly strong, the most popular benchmark was unhedged, and the dollar was the most common base currency. Currency overlay mandates of this specification would be expected to add value!

240

Currency Overlay

11.1.1.2 Watson Wyatt The actuarial and investment consulting firm Watson Wyatt have picked up the mantle left by Brian Strange (Brian left the consulting business, and now works for an overlay manager). The US arm have continued collecting data on currency overlay performance, and have recently released1 their latest findings. Up to the end of 2000, the 5-year median value added of currency overlay managers in their universe who are operating to a USD base and an unhedged benchmark has been 1.32% p.a. The equivalent figure for a 50% hedged benchmark is 0.84%. 11.1.1.3 Frank Russell Frank Russell is an investment consulting firm who have tracked and researched currency overlay managers for longer than any other consultant, and have built a database from contemporary data (rather than ex post) supplied by the managers. They published a report in September 2001 in which they analysed the performance data from 24 currency overlay managers. Their data ran from July 1995 to June 2001, and they found an average outperformance of 1.47% p.a. They analysed the data in somewhat more detail than the Strange Report, and reported value added of 0.94% p.a. for 50% hedged benchmark accounts. Unhedged benchmark accounts showed 2.45% p.a. outperformance, and fully hedged benchmark accounts 0.20% underperformance. This data tends to confirm the suspicion voiced in the Strange Report that time- and mandate-specific factors were distorting the figures in a favourable way. These reports are now uploaded quarterly. 11.1.2 Performance summary The three surveys discussed above, and some other unpublished performance surveys, all point to there being identifiable value added in currency overlay when measured at the level of universe of overlay managers. That this unusual finding is possible is explained by the nature of the FX market – that currency overlay managers have identified and are exploiting one or more persistent market inefficiencies at the expense of the vast majority of the FX market who are constrained in their behaviour and/or are not profit- or price-sensitive.

11.2 WHO LOSES? Elsewhere in the book (Chapter 4, Chapter 9) I have described the mechanics and structure of the foreign exchange market in some detail. If the (small) universe of currency managers has made money on average over nearly 15 years, then the remainder of the FX market, not measured by investment consultants, loses. This is definitional. I have commented elsewhere that the possibility of consistent average outperformance by the universe of overlay managers is a highly unusual position for a group of managers to be in, particularly where there is no real debate about the nature of the benchmark against which they are measured (this is not the ‘outperformance’ of bond managers holding credit measured against government bonds). To be fully intellectually rigorous, however, I should emphasise 1

Brian Hersey and Kurtay Ogunc, Watson Wyatt Presentation, 25 June 2001.

Active Currency Overlay – Evidence of Performance

241

again that there cannot be overall outperformance generated by the currency market in total. Taking all of the participants, including the market-making community, global outperformance is zero. The observation we have made from the surveys, and from the evidence presented in earlier chapters, is that an identifiable class (overlay managers) has made money. This is an observation about style, since the general style that overlay managers adopt (discounting the differences within the overlay universe) is more successful as a trading strategy than at least one other of the groups that make up the total FX market.

12 Implementing Currency Overlay In this chapter I set out a ‘check-list’ for investors considering currency overlay – whether designed to reduce risk, or add value, or both. I also cover some of the practical questions that clients have asked when they establish currency overlay programmes. The answers I give reflect the realpolitik of getting mandates started as well as the theoretical considerations covered in the remainder of the book. Let’s start with the ‘check-list’.

12.1 SUMMARY CHECK-LIST I will raise and answer a number of practical points, which I will summarise in table form in Section 12.1.14. 12.1.1 What mandate type? Currency overlay investors generally come in two categories: 1. Those who wish primarily to reduce their existing international portfolio’s exposure to currency risk. 2. Those who wish to tap a source of excess return from active management of currencies. These two aims may also co-exist, in which case answers to both halves of the question are relevant. (1) Risk reduction: To reach the decision to implement a risk-reduction currency overlay programme, an investor would normally have conducted an asset allocation study, in which ‘hedged equities’ was one of the categories included in the range of assets studied. As a result of the study, the investor would generally be asked to choose (often with the help of an investment consultant) a benchmark hedge ratio for international equities. This is a fundamental asset allocation decision, similar in strategic importance to the domestic:international equities split, and second only to the equity:bond split decision. Many investors do not have a strong instinct for currency and its implications, and therefore they often find this decision difficult to make. The author’s experience is that as a result this decision can be delayed for years, or indeed never properly addressed. It is a minority of investors who systematically analyse the strategic role of currency exposure within their portfolio, and determine a policy, based on these studies, that matches their goals. It is not uncommon for investors to choose an arbitrary benchmark hedge ratio (say 50%) without a full assessment of the implications on the portfolio. While this may appear a little crude, nevertheless it can result in beneficial action where a fuller analysis and more study would result in paralysis and inaction. Active or passive? Once an investor has chosen a strategic position which requires overlay, then he has to decide whether to implement it with an active overlay or a passive overlay. The decision will be determined by two factors. The first (and overriding) is whether the investor

244

Currency Overlay

believes that active currency management can add value. He clearly also needs to believe that he (together with his consultant if he uses one) is capable of choosing an active manager that will add value in the future. In currency overlay this may in practice be much easier than in other active manager searches: as one consultant put it, ‘it is quite hard to find an established currency manager who does not add value over time’. The second factor is whether the investor has a desire for an asymmetrical overlay return pattern. Irrespective of their beliefs about long-term value added, many investors are attracted to the idea of limited negative performance (and unlimited positive performance) of any overlay. Asymmetrical return is also associated with asymmetrical cash flows, so the amount of cash outflow in periods of domestic currency weakness is limited to a predetermined amount rather than dependent on currency movements alone. These asymmetric active mandates are a characteristic of dynamic and option-based styles in particular. Finally, investors may choose to implement a risk-reducing overlay without adopting a hedged benchmark. While this creates a wide tracking error, it does have the merit of an understandable benchmark in which ‘cash-in’ from the overlay means positive performance versus the benchmark, and the positive performance is associated with negative performance of the unhedged international assets. It would be highly unusual, and hard to justify, the adoption of a passive overlay programme with an unhedged benchmark. (2) Value-added: With the increasing length of track records possessed by the main currency overlay firms, most investment consultants and many investors now accept that active currency overlay is capable of generating positive excess returns over long periods of time. As discussed elsewhere in the book (Chapter 4, Chapter 9) this is not the result of chance, nor an unfathomable ‘mystery’. The structural conditions that allow currency market inefficiencies to persist also allow the median currency overlay manager to make money. The design of a pure ‘value-added’ (or ‘alpha’) active currency portfolio is different from a risk-reducing currency overlay, although strictly speaking it is still an overlay since it does not require a capital allocation. The key criteria for a successful ‘alpha’ mandate are: (a) wide latitude given to the manager for cross-hedging and ‘net-short’ positions (i.e. able to exploit any directional movement in any major currency pair); (b) a clear and unambiguous benchmark (zero return is as good as any); (c) a clear expression of the nominal amount of the mandate; and (d) a tracking-error target or limit built into the investment guidelines (see below and Appendix 4). Where the investor has both strategic risk reduction and active value added as goals, then these two, sometimes conflicting designs, have to be knitted together. 12.1.2 For risk-reducing overlay – benchmark hedge ratio The benchmark hedge ratio is, as already discussed (and illustrated) elsewhere in the book (particularly Chapter 6 and Chapter 8), a technically complex area, with some differing academic views, although the views amongst most expert practitioners and consultants are broadly similar, and expressed in this book. A purist approach is to first choose passive or active overlay based on the investor’s beliefs about the continuing ability of currency managers to add value. Then, with that choice in hand, to optimise the allocation of international assets (usually equities in this context) with an additional overlay asset class, as described on p. 134. The optimised portfolio then becomes the strategic asset allocation, and the scale of the active overlay as a percentage of international assets becomes the strategic (benchmark) hedge ratio. Using the values from Table 5.6 (which

Implementing Currency Overlay

245

allows active currency overlay), this would mean a 100% benchmark hedge ratio for all portfolios apart from the most risk-loving – the 15% whole-portfolio volatility target – which would have an 82% benchmark hedge ratio. If passive overlay is the chosen route, then similarly all portfolios apart from the most risk-loving would have a 100% hedged benchmark. However, without active overlay’s value added to maintain its appeal to the optimiser, the removal of the risk constraint sends the benchmark hedge ratio near to zero. However, this purist approach is just that, and I have yet to come across such thorough-going purism in any strategic asset allocation decision. An analytical but realistic approach is to conduct this and other analyses, and then reach a common-sense compromise asset allocation. Most large investors (pension funds, endowments) still favour home bias even though there is no theoretical support for it. They will often measure nominal volatility rather than volatility relative to long-bond returns (reflecting their liabilities), and they will often declare support for one particular asset class over another without a clear description of why they hold that belief. Under these circumstances, a recognition that unmanaged currency exposure gives risk without return will lead such an investor to choose a fully or partially hedged benchmark. This is likely to lead to a benchmark hedge ratio between 50% and 100%, and the final part of the decision will probably depend on the materiality of the international exposure, the maturity of the fund and (for a pension fund) the strength of the sponsoring parent relative to the fund’s liabilities. An intuitive approach would be to accept that currency hedging reduces risk, and therefore that any hedging has merit. Without any further analysis the intuitive investor may wish to minimise the regret from hedging,1 and this will lead to a 50% hedged benchmark. It also provides a symmetrical platform for any active currency overlay that does not allow crosshedging or net-short positions. Finally, many investors are sensitive to the cash flows that arise from currency overlay. The detailed analysis of the costs of managing cash flows in Chapter 6 shows that this sensitivity may not be justified, but nevertheless it exists. High benchmark hedge ratios naturally lead to high potential cash flows from overlay. This may lead some investors to choose lower benchmark hedge ratios than pure risk-reduction considerations would imply, and sometimes unhedged benchmarks (although only for active overlay). 12.1.3 Investment guidelines – active I have included sample investment guidelines in Appendix 4. The key considerations for an active mandate are the scale, the benchmark (discussed above), the restrictions and the targets. The scale is largely self-explanatory, for risk-reducing currency overlay it is normally the marked-to-market valuation of the international assets to be overlaid. It may exclude unhedgeable currencies, although this is usually marginal. Under restrictions come the permitted hedge ratio range (0–100%, 25–75%, 50-150%, etc.), whether cross hedging is permitted (i.e. active positions in currency pairs which exclude the investor’s base currency, and possibly the target currencies), and if cross hedging is permitted, the range, and scale, of currencies that can be traded. Some mandates may have a restriction (i.e. an absolute limit) on downside performance, or on negative cash flows, although this can have the effect of limiting the expected value added of the mandate. Well-designed investment 1 See Gardner, G.W. and Wuilloud, T., The Regret Syndrome in Currency Risk Management: A Closer Look, Russell Research Commentaries, August 1994.

246

Currency Overlay

guidelines give clear instructions in the event of breaches of restrictions, with timetables and consequences. Finally, targets are the key measures on which the success or otherwise of the mandate will be judged. These are typically the tracking error of the actual performance versus the benchmark performance (see Section 7.6), and the average annual excess return (or value added) versus the benchmark. Some investment guidelines can have more complicated targets (‘capture 75% of the increase in value of currencies and limit the losses to 50% of their fall in value’), and the test of these guidelines is how easy they are to monitor and give an unambiguous ‘yes’ or ‘no’ to the question of whether they have been hit or not. In my experience, typical targets for active mandates with a 50% hedged benchmark might be ‘tracking error of 2.5% p.a. with a 1% p.a. excess return’. 12.1.4 Investment guidelines – passive Similar considerations apply to passive mandate investment guidelines as to active ones, but with all the measures writ smaller. The restrictions are likely to be dominant in the operation of the mandate, and the targets should be set on the basis of the manager’s assessment of the impact of the restrictions. Typical targets for a passive mandate with a 100% benchmark hedge ratio might be ‘tracking error of 25 bp p.a. with a –5 bp excess return’ if the benchmark is costless (i.e. based on mid-prices), or ‘tracking error of 25 bp p.a. with a 5 bp excess return’ if the benchmark is based on WM/Reuters left- and right-hand side prices. 12.1.5 Investment guidelines – alpha Investment guidelines for an ‘alpha’ mandate should be much simpler than for a risk-reducing mandate. Guidelines should still have the scale, benchmark, restrictions and targets mentioned above, but they can all be virtually one-liners. Scale can be a nominal amount (preferably in the investor’s base currency) to represent the denominator against which percentages are calculated. Because a currency alpha programme does not require an initial investment, a clear expression of the scale is important. The investment guidelines should make clear how the scale changes over time. Under conventional geometric linking of returns, the scale would reflect the cumulative returns since inception. As an example, if the initial scale of an alpha mandate is $100m, then the scale of the mandate under the given assumed monthly returns would be as in Table 12.1. Note that on p. 84 I explicitly state that monthly returns of overlay alone cannot properly be geometrically linked without including the returns of a ‘carrier asset’. However, alpha is different. The exercise we are conducting here is to express the returns of active currency management with a consistent denominator so the returns can be represented as percentages. If the overlay manager is given these guidelines, then the manager can quite properly rebalance the size of currency positions based on the returns of the currency alpha programme, rather than, in the case of currency overlay, based on the returns of unhedged international assets. Under these circumstances, geometrically linking alpha returns is legitimate. There is another perfectly acceptable approach to take on changes to the scale of the alpha mandate, and that is to have no changes that are not explicit. This would mean that returns from currency alpha would be ‘washed away’, and in the example above, the scale would remain constant at $100m. This approach would mean that the overlay manager would not have to routinely rebalance his alpha process. It would remain legitimate in principle to geometrically

Implementing Currency Overlay

247

Table 12.1 Example of mandate size changes – initial mandate size ($m) = $100m Month 0 1 2 3 4 5 6 7 8 9 10 11 12

Assumed monthly returns 0.21% 0.36% 0.24% −1.72% 1.17% 1.23% 0.49% −0.24% 0.72% 0.08% 0.77% −0.26%

Mandate value ($m) 100.00 100.21 100.58 100.81 99.08 100.25 101.47 101.97 101.72 102.46 102.54 103.33 103.06

link the returns (for example for the purposes of analysis and comparison), but the resulting compound returns would not reflect the live portfolio, which in effect would exhibit simple returns. The most appropriate benchmark for mandates under guidelines similar to these shown above is almost certainly zero. It has the powerful twin benefits of simplicity and transparency. Restrictions on an alpha mandate may be necessary for legal or regulatory reasons (say no net short positions versus the home currency, or limitations on the range of currencies traded). The guidelines will also have to express an annual tracking error, and this could be either framed under this section (implying an absolute tracking error limit), or under the ‘targets’ section, implying no absolute limit. For maximum efficiency, the overlay manager would be given no additional restrictions over those strictly necessary to fulfil legal and regulatory obligations. The most obvious target for the alpha mandate is the annual return target. This has to be expressed in the direct context of the tracking error – they are directly proportional. 12.1.6 Bank FX lines Whatever the type of overlay, an overlay mandate will require the overlay client to have FX lines at a panel of banks. The client may have minimum requirements for credit ratings for the banks, but subject to that restriction, the overlay manager would be expected to suggest the panel based on the particular needs of the client, to negotiate lines between the client and each bank, and to advise the client on the most effective documentation to secure their contractual rights. Where possible, a client’s security is usually best served by entering into IFEMA2 agreements with their counterparties. IFEMA provides a framework for netting (i.e. legally offsetting) matching pairs of forward contracts where one deal reverses the other, and all the other terms, except the contract rate, are the same. This is intended to avoid possible 2 IFEMA (International Foreign Exchange Master Agreement) is a standard set of terms agreed by the British Bankers Association and the US-based Financial Markets Lawyers Group. It was first issued in 1993, and substantially amended and extended in 1997. It is widely accepted in the main foreign exchange centres as ‘standard terms’ governing foreign exchange contracts. A wider version (Foreign Exchange and Options Master Agreement (FEOMA)) covers OTC options as well as forwards.

248

Currency Overlay

‘cherry picking’ of profitable forward contracts by the receiver of one bankrupt contract party, and the repudiation of the loss-making ones. Except under unusual circumstances, a client can expect to be granted FX lines by banks without any requirement for an initial deposit or margin, and without any ‘variation margin’ (see p. 49 for more on margins). This means that the client retains control over the timing of cash payments and receipts from the settlement of forward FX contracts. 12.1.7 Bank contract confirmation The manager and the chosen bank counterparties need to establish a fail-safe mechanism for the confirmation of verbal forward contracts. There are now well-established electronic platforms (for example CMS (formerly FX Match)) which allow both parties to upload deal details into an independent system, and which then provide exception reports where details do not match. The client does not have to be involved in this process, but they need the confidence that it is effective – forward contracts differ from conventional asset purchases in that they are not settled immediately – and of course the delay is the length of the forward contract. Since this is routinely several months, and can be up to 12 months, early and reliable confirmation is at a much greater premium than for conventional cash-settled assets. Where a client’s custodian is asked to record and value outstanding forward contracts, the information that the custodian gets should be after successful confirmation. 12.1.8 Investment management agreement Currency overlay is an investment management assignment, and it should be contractually treated just like a more conventional assignment in equities or bonds. While the investment guidelines for currency overlay are overlay-specific, the investment management agreement can be largely a standard form. 12.1.9 Reporting requirements How currency overlay is reported is critical to the client’s perception of its value, and indeed its performance. The important point here is that the chosen benchmark genuinely reflects the perception of the client (i.e. in answering the question ‘how did the currency overlay manager do?’), and that reporting accurately follows the benchmark decision. As an example: suppose the client chooses a fully hedged benchmark, has given the manager 0–100% hedge ratio leeway (with cross hedging bound by these restrictions too), and the overlay manager is reporting on a period when the domestic currency is weak. The currency overlay will be ‘losing money’, and there will be cash outflows. However, it is likely that the active currency overlay will be adding value versus the benchmark, since only by getting the direction 100% wrong for all currencies could the manager perform as badly as the benchmark. Hence the excess return versus benchmark will be positive, and may be very positive, even while cash haemorrhages. If the currency overlay reports to the client highlight the positive performance, and the client is content with the performance of the manager, then the reporting and the benchmark are congruent. If, on the other hand, the client is uneasy about the cash outflow, or is concerned about the impact of the currency overlay’s performance on the whole portfolio, then the benchmark should probably be reviewed, since it clearly does not represent the ‘neutral’ position of the client. The reporting should then follow the revised benchmark.

Implementing Currency Overlay

249

Two features of currency overlay mean additional reports over and above conventional mandates. The first is cash flow. Cash flow is a (delayed) reflection of the gross performance of the overlay, although not, as we have seen above, of performance versus the benchmark. However, efficient management of the whole portfolio requires that cash flow is forecast as far ahead as possible (although that may not be very far, depending on the overlay manager’s choice of forward contract maturities). These forecasts need to be communicated to the client in cash flow forecast reports. The second is the credit exposure that the client runs when there is strong positive gross performance from the overlay. Under these circumstances, and again depending on the maturity of the forward contracts, the client may wish to know the bank counterparties who owe him substantial amounts of money. This is not academic – where there are concerns with the credit of a large indebted counterparty, it is possible to close out forward contracts and accelerate the settlement of the cash by negotiation with the bank in question. 12.1.10 Periodic cash and contract reconciliation The client will usually choose to use his custodian to record forward contracts undertaken by his overlay manager, and to reconcile these, and the resulting cash flows, with the manager on a regular basis. However, the role of the custodian is a little different in currency overlay from conventional asset management. Forward currency contracts (which are the dominant instrument used by overlay managers) are not initially ‘valuable’ in the way that equities or bonds are. So the conventional role of the custodian to ‘look after’ the securities in a portfolio does not apply in the case of currency overlay. It is perfectly possible, therefore, to not involve the custodian in the overlay record-keeping process, and still retain the security of a third-party check on the accuracy (and existence) of the forward contracts from the bank confirmation process. 12.1.11 Bank contract settlement procedures Except in special circumstances (like manager transition), forward currency contracts will be closed out by the overlay manager prior to maturity. In many cases, the closing-out contract will not be executed until the spot date, which is two days prior to the maturity date (see p. 37 for details). This means that the manager and bank counterparty must then agree the amounts of cash settlement, and the manager is responsible for informing the custodian of the amounts to be paid and received for each relevant value date. The client must be satisfied that the checking, reconciliation and authorisation processes are reliable and secure, although generally speaking the client will not be required to play an active part in this process. 12.1.12 Benchmark calculation Aside from the strategic decision of what the benchmark hedge ratio is to be (for risk-reducing mandates) or what the return benchmark is to be (for alpha mandates), it is clearly important that the client is satisfied that the calculation of the benchmark is accurate and transparent, and ideally independent of the manager. As discussed previously (Table 7.1), there are a number of index providers that produce hedged benchmarks. If one of these fits the mandate for the underlay, then the hedged (or a mix of hedged and unhedged) version provides a ready-made currency overlay benchmark. However, mostly the actual weights of the international assets to be hedged will not exactly match the weights of proprietary indices, and many clients will

250

Currency Overlay

therefore want to develop a bespoke benchmark. There are only three candidates who are generally in a position to design and calculate a bespoke benchmark: the custodian; the client’s investment consultant, and the currency overlay manager. In the author’s experience, custodians have little expertise in currency hedging and currency benchmarks, and their strength lies only in the routine calculations required once the formulae and data sources are established. There is considerable expertise in some of the larger investment consultants, although it tends to reside in specialist currency research units or in particular individuals. Most investment consultants are not geared up to provide regular, systematised monthly reports – so while they may be able to design a benchmark, they probably cannot deliver it in real time. The most detailed expertise, together with systems and processes, to deliver real-time benchmarks lies with currency overlay managers. For many clients, the disadvantage of the loss of independence brought about by using their manager to also calculate the benchmark is overcome by the convenience and accuracy that a knowledgeable and informed manager can bring to the calculation. If the calculation methodology is transparent, and the data source is published and independent, then the loss of independence may be more imaginary than real. 12.1.13 Performance measurement Performance measurement flows naturally out of the benchmark calculation and the reporting of the gross overlay performance. The excess return of the gross overlay performance over the benchmark performance is the bedrock of the performance measurement, and the volatility of that value – tracking error – is the other key measure. With these two values, the progress of the overlay mandate can be monitored, and compared to the targets in the investment guidelines. The principal responsibility for reporting lies with the overlay manager. A well-designed performance report will give the client clear and unambiguous performance reporting. The client, however, may also want the custodian to replicate the key elements of the performance report, both as confirmation of the source numbers (marked-to-market valuation of outstanding forward contracts and cumulative cash flow) and as an independent calculative check. The consultant is likely to be involved in the design-only phase. 12.1.14 Summary check-list Check-list item

Considerations

Decision/action made by

Mandate type

(a) Risk reducing (active or passive); (b) alpha only International exposure and risk aversion Active risk/return positioning

Investor/consultant

For risk-reducing overlay – benchmark hedge ratio Investment guidelines – active Investment guidelines – passive Investment guidelines – alpha Bank FX lines Bank contract confirmation

Tracking error/cash flow considerations Size of mandate/active risk/return positioning Spread of counterparties and credit ratings Fail-safe

Investor/consultant Investor/manager Investor/manager Investor/manager Manager Manager/bank/proprietary electronic confirmation system

Implementing Currency Overlay

251

(cont.) Check-list item

Considerations

Decision/action made by

Investment management agreement Reporting requirements Periodic cash and contract reconciliation Bank contract settlement procedures Benchmark calculation Performance measurement

Commercial, prudential and regulatory considerations To match investor’s needs Fail-safe third-party process

Investor/manager

Secure/accurate/daylight riskminimising Transparent and independent Independent and accurate

Investor/manager Manager/custodian Manager/custodian Consultant/index provider Manager/consultant/custodian

12.2 PRACTICAL QUESTIONS AND ANSWERS Question: I believe that active currency management is no more likely to make money than active equity management. Does this mean I can forget about overlay? Answer: If an investor is not convinced that they can identify an active currency manager, this rules out active overlay and currency alpha. However, whether or not currency can add value, there is a significant reduction in risk available (by adopting a fully or partially hedged benchmark) at almost zero cost for most investors. This would be the adoption of passive overlay. Question: I believe in active currency management. Do I need to have a complete review of my strategic hedge ratio before looking for active currency managers? Answer: No. An investor can exploit currency inefficiencies using currency alpha without changing any strategic benchmarks. Question: How do I manage cash flows from overlay? Answer: For active overlay, limits can be built in to cash outflow. For many investors, this cash requirement will be less than their normal cash allocation, and therefore no special arrangements need to be made. For larger overlay programmes, or for passive programmes, the investor may need to ask the overlay manager and custodian to arrange a process for cash to be realised from (or invested in) low-trading-cost securities (large cap equities, bonds). Alternatively, the investor can hold higher cash allocations with a futures overlay that ‘equitises’ or ‘bondises’ the cash. Question: Can I have overlay without cash flows? Answer: Not for equities. However, it is possible to extend the cash horizon by the use of longer-dated forward contracts. For bonds, however, cash flows can be matched to the interest and bond payments, and in effect become invisible – and this is effected by a currency swap. Question: What fees should I expect to pay for overlay? Answer: Management fees are highly dependent on scale, complexity and other client-specific factors. However, as a guideline, average fees for active overlay tend to be in the 10–15 bp p.a. region (assuming a $500m mandate), and average fees for passive overlay are 3–5 bp p.a. Fees for currency alpha depend on the target level of return, but will generally have a

252

Currency Overlay

management (asset-based) fee component (perhaps 20–25 bp p.a. depending on target return) and a performance-based component (say 20% of new high-water-mark). Question: What exactly does ‘cross hedging’ involve? Answer: Cross hedging is where the manager will undertake forward currency contracts between two currencies, neither of which is the investor’s base currency. This will generally, although not always, mean that the investor increases exposure to one currency (the bought currency) beyond the exposure in the underlying assets. If that is the case, then the hedge ratio for at least one currency will lie outside the 0–100% range. If, despite cross-hedging contracts, hedge ratios of all currencies continue to lie in the 0–100% range, then it could be argued that cross hedging is not really taking place, since that effect could be replicated by doing all deals via the base currency. Question: I have a very low international asset allocation. Can I use overlay? Answer: You can use currency alpha to exploit currency inefficiencies. If your total international exposure is very low (say less than 10% in each asset class), then there is unlikely to be a good case for active or passive risk-reducing currency overlay.

13 Looking Ahead This is a speculative chapter, and since it will reflect my unsubstantiated opinions, it is likely to be of less interest than the earlier chapters with evidence and analysis. It will also be short.

13.1 DEVELOPMENT OF ACTIVE MANAGEMENT STYLES The process by which management styles in any asset discipline become established is a study in itself. A keen observer of investment managers may observe a number of characteristics which seem to be enduring:

r Styles go in fashion – and like fashion, are to some extent ‘returning’. r There is a process of ‘style natural selection’. r Styles become established in any asset class quite early in an asset class’s life; despite small changes at the margins, and the vagaries of fashion, they persist.

r Advances in information processing technology have allowed some quantitative styles to develop, but have not dominated style changes. Currency management styles have conformed to these guidelines quite well. Within a very few years of the advent of currency overlay, three core styles emerged – fundamental, technical and option-based/dynamic hedging. Most managers still recognise these descriptions, and they still give clients and consultants a reasonable guide as to the way a manager constructs a process. Some managers have argued that they no longer fit well into these categories, although my instinct tells me that existing style categories will continue to be pressed into service, even if the fit is not perfect. What about completely new ways of identifying and exploiting currency movements? We could perhaps look at other asset classes to prompt our thoughts. Equity managers use the top-down/bottom-up distinction, and growth/value distinction, to cut styles along two planes. There are also ethical, contrarian/momentum, quantitative and style rotators. They also deal with different classes – large cap, small cap. Hedge funds also have a wide variety of styles unique to them: event-driven (takeovers, distressed situations); global macro (anything and everything), market neutral, convertible arbitrage, and so on. Bond managers, dealing with a very specific asset class, probably have no styles which could transfer to currency. Any opportunism they have displayed in the past in their own currency management of international bond portfolios can be wrapped up in the opportunistic hedge fund style description below. Can any of these concepts apply to currency managers? 13.1.1 Top-down/bottom-up This distinction perhaps resonates with the following in currency: a top-down manager is one who starts the hunt for inefficiencies in the structure of the currency market. The structure applies to all currency pairs, so general statements can be made about all currencies, not just

254

Currency Overlay

specific pairs. ‘Top-down’ currency managers would be students of the market place; they would gather and analyse the nature and motivations of the currency market players, pore over BIS reports when they are issued, lobby for more frequent data about participants and their behaviour, try to assess the effect of changes in the banking industry on the market, and study the changing behaviour of other, identifiable, classes of player such as corporate treasurers, central banks and various investor groups. By contrast, bottom-up managers might be thought of as pure fundamental managers, concerned with the full range of specific factors which impinge on each currency pair. They try to become more knowledgeable about each pair than everyone else in the market (a task so difficult it borders on the impossible). 13.1.2 Growth/value It is a struggle to shoe-horn this distinction into the currency context. Its closest parallel is perhaps the ‘search’ for value in currencies by analysing and exploiting the forward rate bias, since a currency’s interest rate is the closest thing we can get to a value measure. However, the lack of the concept of price/earnings ratio (p/e) in currency, and the role of expectations in the movements of p/e values, means that the analogy largely fails. 13.1.3 Contrarian/momentum I have put contrarian with momentum as two ends of a similar spectrum. They do have a parallel in currencies, which is technical analysis. While contrarians are not formally quantitatively or technically driven, in practice they are mean-reversion supporters. The intellectual argument is a little more sophisticated. A contrarian argues that if ‘everyone’ thinks that stock X or index Y is going up, then that means that all the investors who have the ability to buy it, already have. This means that the only flexibility left to the major participants (who are long) is to sell. This forms the basis for the argument to do the opposite to the ‘herd’. There are very few situations in the currency market when the ‘herd’ is clearly identifiable, and even then usually with the benefit of hindsight. Herds require an element of community and common values/language and culture. These do not exist to any extent in the FX markets. There are many herds, and they all see the world differently. Momentum managers are more overtly technical, but underpinning their approach is to go with the prevailing price trend, and therefore implicitly with the herd. This makes their behaviour mirror the technical trend-follower. Both of these styles resonate to ‘what everyone else is doing’, and so concern themselves with game theory as well as just price movements. The game theory element is less evident in the currency markets than in smaller markets because of the diversity of players, their varied motivations and the opacity of their activities. 13.1.4 Ethical Ethical managers probably do not have a counterpart in currency management. The only parallel that could be drawn is the possible future emergence of a repugnant regime in the current members of the global FX market – currently less than 20 countries. However, one of the characteristics of repugnant regimes is that they tend to deprive their population of freedoms. The continuance of a freely tradeable exchange rate in dictatorial conditions is therefore unlikely, and similarly the emergence of the currency of a repugnant regime into the global market is equally unlikely.

Looking Ahead

255

13.1.5 Hedge fund styles Most of the style categories are inappropriate for currency overlay managers. The only substantive style which has incorporated active currency management is the global macro style. Two events stand out in which hedge funds of this style entered the currency market in a substantial (and price-changing) way. The first is the celebrated case of George Soros and the pound sterling in September 1992. This was, in effect, an event-driven opportunity in which George Soros used the assets of his main hedge fund vehicle, the Quantum Fund, to speculate against the self-imposed lower limit of the pound versus the DEM in the (now defunct) European Exchange Rate Mechanism (ERM). It appears that Soros took a leveraged position of about £10bn against the pound, and this was one of the straws that broke the camel’s back – the UK government capitulating to the market pressure, and (permanent) withdrawal from the ERM. This was a highly publicised win for George Soros, who was rumoured to have made £1–2bn in the Quantum Fund. The second entry of hedge funds into the currency market was less happy. This is the period in 1998 when a group of hedge funds ‘spotted’ what they regarded as an opportunity to take a short position against an already weak yen, and be ‘paid’ for it – the yen ‘carry’ trade. They were ‘paid’ to do this by the interest rate differential between the yen and the USD – yen could be ‘borrowed’ much cheaper than USD could be deposited. The ‘opportunity’ turned out to be a chimera – in two days on 7/8 October 1998 the yen appreciated by about 12%, and the currency market showed real disarray (it was almost a one-way market) – a truly rare event. The trigger for this startling movement was the risk-control processes of the hedge funds trying to close out their rapidly worsening positions all at the same time and in the same direction. Hedge funds collectively lost a great deal of money, and perhaps even more confidence in this debacle. Currency overlay managers were not totally unscathed, but the diversified nature of their portfolios and their diversified styles saved them any serious embarrassment. There are other examples of event-driven ‘opportunities’ being spotted by hedge funds, but many of the lesser known were thoroughly unsuccessful. One of the most important is a series of unsuccessful attacks on the 7.80 HK dollar parity with USD – and the adverse interest rate differential was serious punishment for the lack of success. Will this style enter the currency overlay style universe? I believe that the answer is probably not. Currency overlay’s attraction is its ability to provide more predictable and stable results, from established processes, rather than speculative opportunities with undiversified risks.

13.1.6 Summary on styles None of the existing style descriptors of equity or hedge fund managers fit the needs of currency overlay clients particularly well. Nor do they play to currency overlay managers’ strengths. My own instinct is that technical, fundamental and dynamic hedging styles will (continue to) lose their very distinct characters, and that managers with become less differentiated by these criteria. I believe that all these managers are already exploiting the same inefficiencies, and indeed many of the processes of an overtly fundamental manager may match those of an overtly technical manager. The distinction that I think will become increasingly drawn is, on the one hand, managers who rely on the full rigour of models and processes; and on the other hand, discretionary ‘expert’ managers who are guided by models and indicators, but who take individual decisions

256

Currency Overlay

on the basis of their judgement. We could call these systematic and discretionary. As with all styles, there will be gradations in between the poles, but the meaning is quite clear. Some commentators, and managers, already use one or both of these descriptions to categorise managers. I think this trend will continue.

13.2 NATURAL SELECTION OF OVERLAY MANAGERS According to Mercer, the investment consultant, the first stand-alone currency overlay mandate was signed in April 1985. Prior to this, several (probably around three or four) of the current universe of managers (or their principals) had activities in active currency management. However, none of the contracts under which these activities were undertaken were identifiably currency overlay. By the late 1980s, there were still comparatively few managers offering currency overlay (perhaps six or seven), and there were very few currency overlay mandates. The early 1990s saw rapid growth in the number and scale of overlay mandates, particularly from US pension funds (and some endowments). This encouraged the formation of new overlay managers to service (and stimulate) this demand, and the manager universe grew substantially. A number of the new managers were formed as independent firms by principals who mainly came from investment banks rather than investment managers. Some developed out of a ‘currency desk’ in existing (larger) investment managers. By the late 1990s, the number of managers with $1bn or more under management was probably in the teens. There was a slow-down in the growth of overlay from US institutions after the mid-1990s, but other investing countries, notably Australia, the Netherlands, Switzerland and latterly the UK have continued to create new overlay mandates as their international allocations have risen, and the currency risk has been recognised and dealt with. The currency overlay universe now (in early 2003) probably numbers 40 managers. Many are quite small, and many are also new to overlay. As with all active management populations, there has already been a certain amount of culling of managers, and this will continue. The managers who have been proven most able to survive so far have been those with a well-differentiated style and strong ‘brand’, and those with backing from a large international asset manager, investment bank or custodian. The most vulnerable are the newest, who lack a stand-alone track record, and those with a scale too limited to weather difficult trading conditions. Fees in active currency management are low by the standards of other active management classes. Active currency overlay fees are likely to be in the range 5–50 bp p.a., depending on the scale and complexity of the account. Most active overlay managers will average around 10–15 bp. If we round this to 12 bp, then this is $1.2m in annual fees per $1bn of overlay. Passive mandates would attract fees of 1–10 bp, with most managers averaging passive fees of 2–3 bp for large mandates, although significantly higher for small mandates. There is no evidence that there is any serious upward pressure on fees, particularly in depressed asset return environments. Likewise, the fee levels are probably sufficiently low that there is unlikely to be downward pressure from aggressive pricing from new or growthorientated participants. Table 13.1 gives some very rough guesstimates of the currency overlay market at end-2002 (it is based on international assets shown in Table 2.2). The penetration percentages are my (rounded) estimates; this produces spuriously accurate estimates of overlay mandates in $bn. The accuracy is spurious, and should therefore not be relied on. Most of these mandates will be active overlay, the mix is probably 80:20 active:passive at the time of writing.

Looking Ahead

257

Table 13.1 Overlay geographical penetration analysis Country US UK Switzerland Netherlands Japan Canada Sweden Australia Hong Kong Germany Belgium/France Total

International assets (US$bn)

% Penetration

Overlay mandates (US$bn)

569 291 119 100 64 33 29 23 8 9 8 1251

9% 3% 5% 10% 3% 3% 2% 55% 6% 6% 13% 7%

51.2 8.7 6.0 10.0 1.9 1.0 0.6 12.4 0.5 0.5 1.0 93.7

Source: Record Currency Management Estimates, December 2002.

From this information, we can calculate a very rough approximation of the fee revenues available to the currency overlay manager universe: Assets × Average fee rate = Annual income

= $94bn × [(0.8 × 12 bp) + (0.2 × 3 bp)] = $96m

This means that 40 firms are sharing revenues of $96m, or around $2.5m each if it was equally spread (which it is not). $2.5m is probably below the critical fee income level required to support the research, product development, systems and compliance costs at the minimum level necessary to be a viable manager. This implies that one or more of three things are likely to happen: consolidation of the smaller players into the larger; closure or withdrawal of the smaller players; or significant growth in the overlay market. There is no immediate evidence from the stated intentions of investors that overlay will experience rapid growth over the next few years. Overlay is likely to continue to increase penetration in existing international allocations, and these may well grow with the continuing trend of internationalisation. Countering this, there is some evidence that international allocations in the US have peaked, and that the disappointing return experience of many US investors will cap their desire for more international diversification.

13.2.1 Conflict of interest There is a final point to be made on overlay managers’ income. There are some overlay managers that are part of large investment banks. To the extent that mandates permit them to deal with their own dealing desks, this is a source of customer FX turnover, and therefore profit, to the dealing room. One or two FX dealing rooms have established ‘overlay’ teams based in the dealing rooms, and they are quite explicit about their desire to increase customer turnover. These are not serious overlay managers, and they will always be hugely compromised by the conflict of interest and the dealing culture of their employers.

258

Currency Overlay

More difficult to dismiss are bank subsidiary overlay managers who are set up, and operate out of, an investment management arm of the bank. Many of these are serious and successful overlay managers; it rests with the client to specify the overlay contract in such a way that they can be certain that conflicts of interest do not arise. My recommendation to any client would be to forbid such a manager from dealing with their associated bank. This (harsh) prescription is the only one that can guarantee no conflict.

13.3 EXTENDING THE RANGE OF HEDGING INSTRUMENTS The current range of hedging instruments available to an overlay manager is so liquid, low-cost and flexible that it seems churlish to want more or different. From my experience, the only serious drawback from the customer’s perception of the current range is the regularity, and size, of the cash flows associated with the maturity, and rolling, of forward contracts. Innovation in this area may reduce the hurdles for some investors embarking on overlay, and it is possible that currency swap technology holds the key. The scope of this book does not allow me the space to explore such designs in any more detail.

13.4 WILL INEFFICIENCIES GROW OR SHRINK? Markets are never static creatures, and the FX market is no exception. For most asset markets, the answer to this question is that current inefficiencies will shrink and disappear, and new inefficiencies, currently absent or undreamt of, will appear. The active manager in these markets is therefore like the hunter – equipped to spot, chase and kill the prey wherever it appears. For the FX market, I think the answer is different. The FX market, as I have laboured in this book, is a leviathan that is the clearing-house for a very large volume of trading that is not directly profit-oriented. The secret of how current inefficiencies will develop therefore lies in accurate prediction of how the structure and make-up of the market will change. I see little short-term structural change in the market’s customers. The market’s suppliers, the banks, have undergone radical consolidation over the past 10 years, but this has not markedly changed the clearing or price-setting mechanisms. In the more distant horizon, I see a continuing rise in capital account FX trading (which includes hedge funds and currency overlay managers, and also conventional asset managers, as well as investment banks on behalf of, for example, M&A clients). However, the rate of growth is not likely to be high as long as asset markets remain depressed, and since these sources are a relatively small proportion of the current price-taking customer community at the moment, they will remain so in the mid-term. This leaves the door open for existing inefficiencies to continue to exist, despite their being openly recognised and discussed among the active currency management community. The style of active manager that will thrive in this environment is less like the hunter (responsive, restless, instinctive, action above care) and more like the farmer (planner, order, knowledge, process, care above action). The best farmers are those who possess the properties above, plus an entrepreneurial and intellectually curious streak that leads them to investigate new methods of production, new processes and new machinery. The best overlay managers will possess these qualities too.

Looking Ahead

259

13.4.1 Can outperformance by currency overlay managers continue? The short answer is ‘yes’. The inefficiencies that currency overlay managers exploit are related to the unique structure of the FX market, not to any particular style or ‘view’ held by current participants. However, most textbooks on financial economics will tell you that price inefficiencies are unlikely to be anything more than temporary, since arbitrage will ensure that any ‘free lunch’ is fully exploited until it goes away. Why is this not true in the FX market? It is not true for all the reasons discussed in previous sections: the arbitrage is risky, its existence is not self-evident, it is heavily dependent on specialist FX market expertise, it competes with other opportunities for risk capital, and there is a limited amount of longhorizon risk capital in comparison with the huge volume of capital that flows through the market daily. My prediction is that over, say, the next decade, the apparent value added of the median currency overlay manager may decline, but still remain above zero after costs. This may come about more as a result of the size of the currency overlay manager universe increasing, than by any diminution of the inefficiencies in the FX market. The best, and most consistent, firms will continue to add significant value, but entry of less sophisticated players may dilute the value added of the universe. In my view it is unlikely that any increase in the assets under management in the active overlay universe will by itself be sufficient to significantly reduce or eliminate the persistent inefficiencies in the FX market.

References/Useful Reading AIMR Benchmarks and Performance Attribution Subcommittee Report, John C. Stannard et al., August 1998. AIMR Conference Proceedings, Currency Risk in Investment Portfolios, 1998. Aitchison, J. and Brown, J.A.C., The Log-normal Distribution, Cambridge University Press, 1963. Bernstein, W.J. and Wilkinson, D., ‘Diversification, rebalancing, and the geometric mean frontier’, Efficient Frontier (www.effisols.com), January 1998. Black, F. and Scholes, M., ‘The pricing of options and other corporate liabilities’, Journal of Political Economy 81 (May/June 1973) 637–659. Black, F., ‘Living up to the model’, Risk Magazine, March 1990. Black, F., ‘Universal hedging’, Financial Analysts Journal, July–August 1989. Brigham, E.F. and Houston, J.F., Fundamentals of Financial Management, The Dryden Press, 8th edn, 1998. Clowes, M.J., The Money Flood, John Wiley & Sons, 2000. Econometric Methods, Johnston and Dinardo, McGraw-Hill, 4th edn, 1996. Fama, E.F., ‘Market efficiency, long-term returns and behavioural finance’, Journal of Financial Economics 49 (1998) 283–306. Fama, E.F., ‘Efficient capital markets’, Journal of Finance 46 (Dec 1991) 1575–1617. Fama, E.F., ‘Random walks in stock market prices’, Financial Analysts Journal, September–October 1965. Fitzgerald, M., Financial Options, Euromoney Publications, 1987. Gardner, G.W. and Wuilloud, T., ‘The Regret Syndrome in Currency Risk Management: A Closer Look’, Russell Research Commentaries, August 1994. Gartland, W.J. and Letica, N.C., ‘The basics of interest-rate options’, chapter 59 in Fabozzi, F.J. (ed.), The Handbook of Fixed Income Securities, McGraw-Hill, 5th edn, 1997. Haugen, R.A., Modern Investment Theory, Prentice Hall, 1993. Jorion, P., ‘Mean/variance analysis of currency overlays’, Financial Analysts Journal, May–June 1994. JP Morgan/Reuters, ‘RiskMetricsTM – Technical Document’, 4th edn, 1996. Karnosky and Singer, ‘Global Asset Management and Performance Attribution’, Research Foundation of The Institute of Chartered Financial Analysts, February 1994. MacDonald and Stein (eds), Equilibrium Exchange Rates, Kluwer Academic, 1999. Markowitz, H.M., ‘Portfolio selection’, Journal of Finance 7 (1952) 77–91. McCutcheon, J.J. and Scott, W.F., An Introduction to the Mathematics of Finance, ButterworthHeinemann, 1986. Nesbitt, S.L., ‘Currency hedging rules for plan sponsors’, Financial Analysts Journal, March–April 1991. Perold, A.F. and Schulman, E.C., ‘The free lunch in currency hedging: implications for investment policy and performance standards’, Financial Analysts Journal, May–June 1988. Rogoff, K., ‘The purchasing power parity puzzle’, Journal of Economic Literature 34(2) (1996). Rosenberg, M., Currency Forecasting: A Guide to Fundamental and Technical Models of Exchange Rate Forecasting, Richard D. Irwin & Co, 1996.

262

References

Rosenberg, M., The Deutsche Bank Guide to Exchange Rate Determination, May 2002. Sarno, L. and Taylor, M., ‘Official intervention in the foreign exchange market: is it effective, and if so, how does it work?’, Journal of Economic Literature 34(3) (2001). Sharpe, W., ‘Asset allocation: management style and performance measurement’, Journal of Portfolio Management, Winter 1992. Siegel, J.J., ‘Risk interest rates and the forward exchange’, Quarterly Journal of Economics 89 (1972) 173–175. Taylor, A.M., ‘A Century of Purchasing-Power Parity’, NBER Working Paper No. w8012, November 2000. Tobin, J., ‘A proposal for international monetary reform’, Eastern Economic Journal 4 (1978) 153–159.

Appendices APPENDIX 1 – BOUNDARY CONDITIONS FOR FORWARD ARBITRAGE The following sets out the maths that applies in each case: Case 1 The ‘no spread’ world – sterling investor replacing a sterling deposit with a hedged dollar deposit. Case 2 Real markets – sterling investor replacing a sterling deposit with a hedged dollar deposit. Case 3 Real markets – dollar investor replacing a dollar deposit with a hedged sterling deposit. Case 4 Real markets – borrowing one currency to lend the other, the outer bounds of the arbitrage. Case 1 – No spreads Invest in sterling direct: (1) Invest an amount of sterling £DEP (2) Interest = £DEP × £LIMID/12 (3) Total £ value 1 month £DEP[1 + (£LIMID/12)] = £RET Invest via the dollar: (4) Invest £DEP (5) Convert to $ = £DEP × $£MIDSPOT = $DEP where $£MIDSPOT is expressed as dollars/sterling: (6) Interest = $DEP × $LIMID/12 (7) Total $ value 1 month $DEP[1 + ($LIMID/12)] = $RET (8) Total £ value 1 month $RET/ $£MIDFWD The arbitrage is perfect where the total £ value 1 month is equal from both investing routes, i.e. (3) = (8), or: (9) £DEP[1 + (£LIMID/12)] = $RET/$£MIDFWD Substituting from (7): (10) £DEP[1 + (£LIMID/12)] = {$DEP[1 + ($LIMID/12)]}/$£MIDFWD and rearranging to find the forward rate ($£MIDFWD): (11) $£MIDFWD = {$DEP[1 + ($LIMID/12)]} / {£DEP[1 + (£LIMID/12)]} Substituting from (5):

264

Currency Overlay

$£MIDFWD = {(£DEP × $£MIDSPOT) × [1 + ($LIMID/12)]}/ {£DEP[1 + (£LIMID/12)]} and simplifying: (13) $£MIDFWD = $£MIDSPOT × [1 + ($LIMID/12)] / [1 + (£LIMID/12)] Or in words, the forward rate is equal to the spot times the ratio of 1 + interest rates (the ratio has the numerator interest rate on top). We can substitute example values as follows: (12)

$£MIDSPOT = 1.5346 £LIMID = 4.9688%

$LIMID = 2.0938%

which gives the forward rate as follows: (14) $£MIDFWD = 1.5346 × [1+ (2.0938%/12)]/[1 + (4.9688%/12)] = 1.5346 × (1.001745/1.004141) = 1.53094 and therefore the ‘FX swap’ expressed in ‘FX points’ (1 point = 1/10 000th of $1) at mid-prices is: (15) 1.53094 − 1.5346 = −36.6 points Case 2 – Sterling investor with spreads Invest in sterling direct: (1) Invest an amount of sterling £DEP (2) Interest = £DEP × £LIBID/12 (3) Total £ value 1 month £DEP[1 + (£LIBID/12)] = £RET Invest via the dollar: (4) Invest £DEP (5) Convert to $ = £DEP × $£RHSSPOT* = $DEP where $£RHSSPOT is expressed as dollars/sterling: (6) Interest = $DEP × $LIBID/12 (7) Total $ value 1 month $DEP[1 + ($LIBID/12)] = $RET (8) Total £ value 1 month $RET/$£RHSFWD* One further piece of information is needed, and it refers to $£RHSSPOT and $£RHSFWD. The way in which the forward market is priced is by forward FX points, so, for example, the forward discount of the pound in the previous example might be quoted as −37/−36 (i.e. a ‘one point’ spread). This means that for the deal contemplated above, the bank will sell the customer dollars at the spot rate, and simultaneously buy it back from them for 1-month maturity at 36 points better (better to the investor) rate (i.e. lower number). The bank will not charge a spot spread, because no spot risk is being taken by them. But in this example, the amount that this investor wants to sell forward is larger (by the interest received) than the amount they are buying, so there will be a very small outright forward deal, which will not only attract a full spot and forward spread, but will be used by the bank as the spot rate off which they will price the whole deal (and this will be the right-hand side (RHS)). In a pure FX swap (i.e. the two amounts of dollar being identical), the bank is indifferent to the spot rate used, and might accept LHS, mid or RHS without preference. This is why I have used $£RHSSPOT as the spot conversion rate, whereas normally the investor would buy dollars at $£LHSSPOT. From the foregoing, we can define the forward rate in this case as follows:

Appendices

265

(9)

Outright forward rate: $£RHSFWD = $£RHSSPOT + ($£RHSSWAP /10 000) where $£RHSSWAP is the RHS swap points expressed in FX points. Using exactly the same algebra as in the previous case, we can derive the forward price which is just ‘no arbitrage’. So mirroring the previous equation (13): (10) $£RHSFWD = $£RHSSPOT × [1 + ($LIBID/12)]/[1 + (£LIBID/12)] We can substitute example values drawn from market data as follows: $£LHSSPOT = 1.5345

$£MIDSPOT = 1.5346 $£RHSSPOT = 1.5347 £LIBOR = 5.0000%

$LIBOR = 2.1250% $LIMID = 2.0938% $LIBID = 2.0625%

£LIMID = 4.9688% £LIBID = 4.9375%

(11)

$£RHSFWD = 1.5347 × [1+ (2.0625%/12)] / [1 + (4.9375%/12)] = 1.5347 × (1.001719/1.004115) = 1.53104 and therefore swap points: (12) 1.53104 − 1.5347 = −36.6 points which is the same value as the mid-price calculation. So, perhaps a little surprisingly, the arbitrage kicks in instantly – if the UK investor is ever offered better than −36.6 points (i.e. a larger number), the US investment route will risklessly make him more money than an investment in sterling. It is worth noting that the spot bid/offer spread is irrelevant for this arbitrage – even if interest rates are very high, and, say, the amount of the sale of dollars was 10% more than the amount of the purchase. Case 3 – Dollar investor with spreads Invest in the dollar direct: (1) Invest an amount of dollars $DEP (2) Interest = $DEP × $LIBID/12 (3) Total £ value 1 month $DEP[1 + ($LIBID/12)] = $RET Invest via sterling: (4) Invest $DEP (5) Convert to £ = $DEP/$£LHSSPOT = £DEP where $£LHSSPOT is expressed as dollars/sterling: (6) Interest = £DEP × £LIBID/12 (7) Total £ value 1 month £DEP[1 + (£LIBID/12)] = £RET (8) Total $ value 1 month £RET × $£LHSFWD The algebra from here is similar to the previous cases, but we have reversed the direction of the foreign exchange transaction, so we divide where previously we multiplied and vice versa: (9) $DEP[1 + ($LIBID/12)] = £RET × $£LHSFWD Substituting from (7): (10) $DEP[1 + ($LIBID/12)] = £DEP[1 + (£LIBID/12)] × $£LHSFWD and rearranging to find the forward rate ($£LHSFWD): (11) $£LHSFWD = {$DEP[1 + ($LIBID/12)]}/{£DEP[1 + (£LIBID/12)]}

266

Currency Overlay

Substituting from (5): (12) $£LHSFWD = {$DEP[1 + ($LIBID/12)]}/ {($DEP / $£LHSSPOT) × [1 + (£LIBID/12)]} Simplifying and rearranging: (13) $£LHSFWD = [1 + ($LIBID/12)] / {(1/ $£LHSSPOT) × [1 + (£LIBID/12)]} (14) $£LHSFWD = $£LHSSPOT × [1 + ($LIBID/12)]/[1 + (£LIBID/12)] Or in words, the forward rate is equal to the spot times the ratio of 1+interest rates (the ratio has the numerator interest rate on top). This is the same algebra as Cases 1 and 2 – just the LHS and RHS are different. So it does not matter whether the investor who exploits these arbitrage opportunities is dollar-based or sterling-based. Substituting with values above: (15) $£LHSFWD = 1.5345 × [1+ (2.0625%/12)] / [1 + (4.9375%/12)] = 1.5345 × (1.001719 / 1.004115) = 1.53084 and therefore swap points: (16) 1.53084 − 1.5345 = −36.6 points which is again the same value as the mid-price calculation. This means that the dollar-based arbitrageur will make riskless gains if the swap points are numerically less than this −36.6 (i.e. if they were 36). So there are zero bounds for this arbitrage, even in the real world (since the sterling investor will make money if the swap points are numerically higher than −36.6, ( i.e. 37). This is surprising – normally the transactions costs of the real world mean that arbitrage has a ‘neutral’ area where it is profitable for no-one.

Case 4 – Borrowing one currency to lend the other The final case we will examine is where an arbitrageur is not a natural lender or borrower. Here, he will have to borrow one currency and lend another, which will mean that he will have to overcome the interest rate bid/offer spreads to be able to exploit the arbitrage. The base currency of the investor is irrelevant (and even could easily be a third currency) – the key is the boundary forward points beyond which the arbitrage is profitable. The principle of this arbitrage is simple: let’s start with an investor borrowing sterling to invest in dollars. The maths is identical to Case 2, except that equation (3) in case 2 will have a £LIBOR term in, not a £LIBID. This works through to the following equation (matching (10) in Case 2): (1) $£RHSFWD = $£RHSSPOT × [1 + ($LIBID/12)] / [1 + (£LIBOR/12)] Substituting values: (2) $£RHSFWD = 1.5347 × [1+ (2.0625%/12)] / [1 + (5.0000%/12)] = 1.5347 × (1.001719 / 1.004167) = 1.53096 and therefore swap points: (3) 1.53096 − 1.5347 = −37.4 points So if the RHS of the swap points rises higher than −37.4 (e.g. 38), then the borrowing sterling/ lending dollars arbitrage is profitable. A similar boundary can be established for the LHS of the swap. The maths is identical to Case 3, except that equation (3) in Case 3 will have a $LIBOR term in, not a $LIBID. This works through to the following equation (matching (14) in Case 3):

Appendices

267

(4) $£LHSFWD = $£LHSSPOT × [1 + ($LIBOR/12)] / [1 + (£LIBID/12)] Substituting values: (5) $£LHSFWD = 1.5345 × [1+ (2.1250%/12)] / [1 + (4.9375%/12)] = 1.5345 × (1.001771 / 1.004115) = 1.53092 and therefore swap points: (6) 1.53092 − 1.5345 = −35.8 points So if the LHS of the swap points falls below −35.8 (e.g. 35), then the borrowing dollars/lending sterling arbitrage is profitable. These arbitrage possibilities make currency forward rates virtually always a pure mathematical result of relative interest rates. The question of whether this always means that interest rates determine forward rates is a sophisticated question, the answer to which involves the relative weights of money following domestic interest rates or forward rates. Most of the time domestic rates dominate, but sometimes it is clear that forward rate expectations dominate. This has been most visible recently in the attacks against the French franc peg versus DEM (1982 and 1983), sterling peg versus DEM (1992), French franc versus DEM (1993) and Hong Kong dollar versus US dollar (1998), when domestic interest rates became distorted by views of the future exchange rate. At its most extreme, French franc ‘Euro’ (i.e. offshore) overnight rates went to 3000% p.a., equivalent to an 8% devaluation the next day. This was only possible because of exchange control, which kept domestic French rates to about 20% p.a.

APPENDIX 2 – LOGNORMAL RETURNS Setting out the problem In practical fund management, we are often faced with time series of returns (or asset values), and need to summarise the characteristics of these series with annualised returns and volatility. Let’s take a 1-year period of total monthly returns from some conventional asset (say an equity; Table A2.1) and ask the simplest questions: what are the observed annualised return and SD?1 Table A2.1 Period 1 2 3 4 5 6 7 8 9 10 11 12 1

I use volatility and SD interchangeably.

Returns 2.4142% 4.9287% 1.3992% 2.5029% −1.4179% 6.6249% 5.5299% 4.4657% 2.9206% 0.1610% −3.5356% −3.6404%

268

Currency Overlay

Return The compound return is the product (less 1) of (1 + period return) for the 12 periods. The reader might like to ascertain himself that this is 24.02%. We do not need to do any explicit annualising because the 12 periods are one year’s worth. There are two other obvious arithmetic or hybrid methods, which I mention to dismiss: Arithmetic average = Average of monthly returns × 12 = 22.35%

Hybrid = {[(Average of monthly returns) + 1] ˆ12} − 1 = 24.79%2

I dismiss these measures because they fail our basic definition; the asset value after one year of the asset above is $124.02 for a starting value of $100. Note that in this example the arithmetic average return is lower than the compound; this is commonly (but not universally) the case for positive return series when annualising monthly data over only one year. However, when we annualise longer series of data, the effect is reversed, and compound returns are typically lower than arithmetic. This effect is emphasised by high volatility series, a phenomenon often referred to as ‘volatility drag’.3 Volatility This is a much less straightforward problem. Firstly, we should note that the answers in the returns section above were all given without mentioning logs.4 In volatility calculations, however, we will find that logs are essential. Let’s go back to first principles. The series in Table A2.1 was generated by a lognormal random generator model. For each month, the model selected a random point on a normal distribution5 with mean R and volatility V . In this model, R = 0.007943 and V = 0.040346. We will come back to why these values were chosen later. The randomly generated values for the 12 months were as given in Table A2.2. Note that while I have labelled these values ‘log values’ there is nothing Table A2.2 Period

Log values

1 2 3 4 5 6 7 8 9 10 11 12 Sum of logs

0.023855 0.048110 0.013895 0.024721 −0.014281 0.064147 0.053824 0.043688 0.028788 0.001609 −0.035997 −0.037083 0.215276

√ I use ‘ˆ’ to mean ‘raise to the power of’. This means x and x ˆ 0.5 are identical. McCutcheon and Scott (1986) show the mathematics of volatility drag on p. 293. 4 Throughout this appendix ‘log x’ or ‘ln x’ means log base e of x, and ‘exponential of x’ or ‘exp(x)’ means ex . e = 2.7182818 . . . . 5 In Microsoft Excel this can be calculated by using =NORMINV(RAND(),R,V), where R = mean return and V = volatility. 2

3

Appendices

269

particularly ‘log’ about them at this stage; they are normally distributed, for example. However, to create a lognormal distribution of returns, we take these numbers and treat them as if they were logs. This means that to ‘apply’ them to asset values over time we take the exponential of each of these values, and multiply the end-month asset values by these exponential values. Table A2.3 shows this mathematical process clearly in each of its stages. We can recognise column (e) as our series of returns.6 We also note that returns = exp(logs) − 1. At this juncture we can also succinctly show that compounding returns using geometric linking exactly matches the lognormal model. The annualised return using compounding is 24.02%. The sum of the logs in column (b) is 0.21527. Exp(0.21527) − 1 = 24.02%. For longer periods than a year (say from Table A2.4 later) the reader might like to take the average monthly log return and multiply by 12. The exp()−1 result will be identical to the annualised compound return produced by geometric linking [={(asset value at end/asset value at start) f /n }−1, where f = data frequency per year and n = no of periods]. Table A2.3 (a) Period

1 2 3 4 5 6 7 8 9 10 11 12 Sum of logs

(b) Random log value

(c) Exponential value [=exp(b)]

(d) Asset value ($) [=c(t)d(t−1)]

0.023855 0.048110 0.013895 0.024721 −0.014281 0.064147 0.053824 0.043688 0.028788 0.001609 −0.035997 −0.037083 0.215276

1.024142 1.049287 1.013992 1.025029 0.985821 1.066249 1.055299 1.044657 1.029206 1.001610 0.964644 0.963596

100 102.41 107.46 108.97 111.69 110.11 117.40 123.90 129.43 133.21 133.42 128.71 124.02

(e) Returns [={d(t)/d(t−1)}−1] or [c−1]% 2.41% 4.93% 1.40% 2.50% −1.42% 6.62% 5.53% 4.47% 2.92% 0.16% −3.54% −3.64%

Turning again to volatility, we know that the SD of the population from which these returns were drawn was 0.040346. However, we also know that they were transformed by the exponential function to reach percentage returns. So to get to the volatility of this return series, we could go through the same transformation process. Firstly, we take the SD of column (b).7 The reader might like to ascertain for himself that it is 0.033646. But this is the volatility of the 1-month log values, whereas we want the annualised volatility. The Central Limit Theorem 6 We could have done it a different way: we could have taken the log of the initial asset value of $100 [ln($100) = ln$4.60517] and added the numbers in column (b) in sequential order to give us a ln(asset value) each month. Log asset values are not particularly intuitive, but once exponentiated they come out fine. Using this method, asset value at the end of the year is: exp(4.60517+0.215276) = exp(4.820446)  = 124.02. n 

(li −l )2 where li = ith period log value, l = average value of l in the sample, and n = number of observations in the SD = i=1n−1 sample.√There is some debate about the denominator being n or n − 1. Using n − 1 gives an unbiased estimate of variance (but not SD = variance). However, I will use it as most investment management users do. See however the AIMR section. 7

270

Currency Overlay

tells us that the SD of the mean√of a sample size of n independent observations from a normal population with SD= V is V / n. However, we are not averaging here, we are summing √– we want the sum of√the 12 observations to make up the year’s return. We therefore get nV / n, or rearranging, V n. So the annualised SD of this series is 0.033646 × (12 ˆ 0.5) = 0.116553. Note that we can only sum returns if they are log returns; if we try to √ sum percentage returns to annualise, we are being inconsistent with geometric linking. So nV/ n is only valid for log returns, not percentage returns. The users of the conventional percentage SD calculation have probably not realised this. But this value is in logs. To convert to percentages as defined, we have to transform using the exponential function: SD = exp(0.116553) −1 = 12.36%. So you would think that, following this logic, the volatility of this series was 12.36%. I should just say that the population SD of 0.040346 is derived √ from wanting to generate a 15% annualised volatility series. If V = 0.040346, then exp( 12V)−1 = 15%. Industry Practice Faced with the return series in Table A2.1, I suspect when asked to calculate the sample√SD, most industry participants would take a straight SD of the percentages, and multiply by 12. This comes to 11.79%.8 There are a number of points to make here. The first is that if returns are lognormal, then an SD cannot be calculated as if the returns are additive (i.e. as the SD of the % period returns). Just imagine a very high volatility investment of SD = 50%. What happens if the returns are negative 3 SDs? Does the value of the investment go to less than zero? Figure A2.1 shows the range of percentage changes for a series with a mean annual return of 10% and SD of 15% for (a) lognormally calculated percentage changes for −4 to +4 SDs (the SD applied to the log of the % returns and converted back to %) and (b) additive percentages for the same range (the SD applied directly to the % returns). It shows quite clearly that additive percentages underestimate the risk of high returns, and overestimate the risk of low returns. In short, additive percentages show no convexity.9 We should note that the 1 SD positive value for the log series is 26.5% = exp[ln(1.1) + ln(1.15)] − 1, whereas for the additive series it is 25% (10% + 15%). This is my guess as to industry practice: what evidence do we have for this?

MSCI MSCI, the index provider, has published very comprehensive index methodology10 and, in summary, it uses additive returns to calculate the annualised return of the SD. Their formula is:    n (TR j − TR)2  √  j=1 × 12 Annualised volatility = (A2.1) n−1 8

In Microsoft Excel terms =STDEV(Series)∗ (12 ˆ 0.5). Convexity in returns (and asset values) is characteristic of lognormal processes, and for the mathematically-minded the second differential of the curve in Figure A2.1 is positive for the convex process, zero for the additive process. 10 www.msci.com/risk/index.html 9

Appendices

271

100.0% 80.0%

Logs

Returns %

60.0%

Additive percentages

40.0% 20.0% Mean 0.0%

−20.0% −40.0% −60.0% 4 SDs

3 SDs

2 SDs

1 SD

Mean

1 SD

2 SDs

3 SDs

4 SDs

Normalised range of returns Figure A2.1 Lognormal return distribution for a series with mean 10% and SD 15%

where TR j symbolises the monthly total return, TR is the average total return and n is the number of data points used in the calculation. MSCI’s TR value is a percentage monthly return; they spell out in great detail their methodology for annualising TR, and they do so correctly using geometric linking and annualising by raising the product of (1 +TR) to the power of 12/m, where m is the number of months to be annualised. All of which is fine, and fits exactly with log methodology as described above. It is all the more worrying then that they abandon their log methodology when it comes to the calculation formula above for the SD of the series, and calculate the SD of the % TR j , rather than via ln(1 + TR j ). Their formula therefore treats returns as if they were additive, and incorporates no convexity. Interestingly, MSCI go on in their methodology document to show a Value at Risk (VAR) calculation as follows: VAR = (1 − e[−1.645σ ] ) × 100

(A2.2)

where 1.645 implies a 95% confidence level, σ is the monthly standard deviation (volatility) of the monthly total returns and e is a constant term having an approximate value of 2.718. Assume Poland has a VAR of $15. This implies that for every $100 invested in Poland, there is a 5% chance of losing $15 or more in a given month. This is particularly odd, because the introduction of the e-power function (or exp()) would be absolutely correct if the SD was an SD of log returns, but is absolutely mathematically wrong for an SD of total return %. Rewriting the formula (A2.2) using the exp function: VAR = [1 − exp(−1.645SD)] × 100

(A2.3)

272

Currency Overlay

We could expand this formula to: VAR = [1 − exp(−1.645{SD(exp(ln(x)) − 1)})] × 100

(A2.4)

where x is normally distributed, because the TR are already exp() of the log returns, so we have a ‘double exponentiation’. We also have the problem of pro-rata-ing up from monthly to annual. This will be addressed later. If MSCI were to stick with the logic of their additive SD calculation, then the VAR would be: VAR = [1.645SD] × 100

(A2.5)

which would be 19.39 from our series above. The irony is that with the same data that MSCI have, we can calculate the correct result. We calculate an SD of the log returns, which are ln(1+T R j ). I will call this SD(ln). VAR is then: VAR = [1 − exp(−1.645{SD(ln)})] × 100

(A2.6)

Using the series in Table A2.3, the correct calculation of the VAR is 17.45, whereas MSCI’s calculation would be 17.63. Do not be fooled by this relatively small error; it is not stable and it could be much larger depending on the data set and the annualisation process. More of this later. AIMR What about the Association for Investment Management and Research (AIMR)? AIMR is the US performance reporting standards setter, and is the leading partner in the new global performance standard – GIPS. AIMR has published its preferred methodology for the calculation of SD in Appendix B of the 1993 Performance Presentation Standards. It describes SD formulae in two places (external and internal risk). The basic formulae are the same, and are as follows. ‘Standard deviation of portfolio performance over time (S p ) is a measure of volatility. . . calculated as follows:    n  (CASSETi − M E AN (CASSET ))2  i=1 (A2.7) Sp = n

where CASSETi is the asset weighted composite return in the ith time period, and n is the number of periods in the study.’

AIMR do not mention the lognormality issue at all, although like MSCI they imply that returns are lognormal by their insistence on geometric linking for returns over time. Like MSCI, they too see the SD of percentage returns as the appropriate measure. By way of an aside, AIMR take some time to justify the use of n rather than n − 1 as the denominator. It is a statistical debate at a tangent to the present issue – so I will ignore it.

Appendices

273

Textbooks and academic literature There is a huge volume of textbook and academic literature from which to draw samples of common usage. I have picked two textbooks at random: Haugen (1993) and Brigham and Houston (1998). The latter is particularly relevant because it is a recommended text for the CFA – the US (and now worldwide) industry standard investment management qualification. Both of these books unequivocally calculate SDs of investment returns using the (incorrect) methodology shown earlier, or generically:    n  (ri − r )2  i=1 SD = (A2.8) n−1 where ri = ith period return in %, r = average value of r , and n = number of observations in the sample. However, a number of commentators remark (in passing) that log returns are the ‘correct’ measure, but then go on to say that ‘percentages are good enough’. An example of this comes from Gartland and Letica (1997) in the bond market literature: ‘The intuitive approach to calculating a percentage volatility is to find the standard deviation of daily returns, assuming a normal distribution. This approach is equivalent to the lognormal assumption as long as the distribution can be characterized as being equally normal and lognormal and the changes in prices are taken on a small interval, such as daily.’ These comments seem to abound – that percentages are fine as long as the volatilities and/or time period are small. Of course this is wrong. A daily SD of 1% (and most commentators would say that 1% is ‘reasonably small’) annualises to 17.48% if correctly calculated via logs, and to 16.19% if incorrectly calculated direct. A 2 SD movement upwards is 38.01% (if calculated correctly via logs), and only 32.37% if calculated incorrectly with direct percentages. These are not second-order errors. What about the founders of modern portfolio theory? Markowitz (1952), a model of clarity of thought struggling against computational odds, does discuss logs in relation to compound returns and their geometric mean. However, he is principally concerned with his emerging theory of efficiency, and the mathematics of combining stocks to form efficient portfolios. Because period returns are easy to observe and intuitively appealing, he chooses to stick to returns as his core ‘random variable’. While recognising the inadequacy of adding returns over time rather than compounding, he seeks to provide a quadratic formula for approximating the compound return from return data rather than effecting a fundamental transformation of his return data into logs. It is quite possible that this approximation (adopted, I suspect, largely because of the computational difficulties of wholesale transformations into and out of logs) has carried on down the years just because no one has thought hard about an alternative. So in the end, Markowitz calculates the SD of returns directly, not via logs. Investment banks and option traders One final industry sector has not yet been considered. This is one where the exact calculation of volatilities for the pricing of options and other exotic derivatives is critical, since large volumes of trades are transacted on the basis of these calculations every day. It consists of the options desks of the major investment banks.

274

Currency Overlay

From a brief survey of custom, practice and use in this sector,11 it is clear that logs are universally used for SD calculations, but that, bizarrely, the logs are not converted back to percentages, but left as logs with a ‘%’ sign on the end. This is extremely confusing, but as long as this is consistently and universally applied, no errors will arise. In particular, annualising will be intuitive, and the errors described in the section below will not arise. However, if this convention is universally applied to returns and SDs, then a return of 10% under this convention would turn $100 into $110.52 at the end of one year. This is because the 10% here is a log, and the asset value will be multiplied by exp(0.1) to get the next value. Exp(0.1) is 1.1052. This is equivalent to continuously compounding 10% over infinitesimal time. While mathematical models find this easy to do, investors find continuously compounding hard to handle and even harder to re-create with data! This treatment is confirmed in the textbooks written for the investment banking community. Fitzgerald (1987), writing about option pricing, describes a fully consistent log calculation of SD, but right at the end he says: √ ‘Annual volatility = 0.003625 × 365 = 0.069 or 6.9%’ His use of % is confusing, as it implies that it refers to returns, when in fact it refers to ln(1 + r ) where r = returns. It is also interesting that he uses 365 as the annualising factor. So the investment banking conventions violate our investment management convention of 10% return meaning you have $110 at the year-end from your $100. I see nothing wrong in the investment banking community continuing to use logs as the basis of everything they do, but I do not believe their usage would translate well to the investment management community. Finally, as an aside, I am not sure that many participants in the investment banking community really understand that they are dealing with logs, particularly since everything they do is expressed as x%. I see logs as a particular language (call it Latin), and percentages as another language (call it English). There is an exact translation process to convert one to the other. But for option traders to describe an implied volatility as ‘12%’, when this is in fact a log value of 0.12 (which would translate to a return volatility of exp(0.12)−1 = 12.75%) in my view runs the risk of using Latin words that look like English words but have a different meaning. I will always use the convention of x% when referring to percentages, and y.yyy when referring to logs. It is possible that the reader is already convinced by the theoretical arguments presented above. However, there may be a strong element of ‘so what?’ from readers who see the errors I have demonstrated to be rather esoteric, and the SD of % returns to be a reasonable approximation of the more precise log-based SD. I hope that the following example will convince waverers. Let us suppose we are dealing with a very volatile stock – say an Internet stock – whose volatility we wish to measure. The monthly returns for 2 years are given in Table A2.4. This stock has observed annualised volatilities as follows: 1. Correct (ln) calculation 2. SD on return percentages 3. SD of log

SD= 109.03% SD = 78.60% SD = 0.7373

11 See, for example, JP Morgan (1996). There is a very good analysis of the relative merits of percentage and log returns in section 4.1. They choose log returns for all their calculations, but they do note that this is not as convenient as percentages for portfolio cross-sectional analysis. They do not ‘re-translate’ log returns back into percentages, but nevertheless they use ‘x.x x%’ notation.

Appendices

275

I suspect that the differences between values 1 and 2 is large enough to worry the most hardened ‘approximator’, and it of course brings into question the whole concept of percentage changes that can in theory be larger than 100%. Raw percentages cannot cope with this level of volatility; by contrast, log returns can cope with any level of volatility. The first thing that any modeller will notice when faced with annualised volatility of 109% (or indeed 78%) is that there seems to be a high chance each year that the value of the stock will go negative. Since this is impossible in a limited liability company’s equity, he is forced to override at zero value, or put in some other ‘fudge factor’ to prevent this possibility. Table A2.4 Period 1 2 3 4 5 6 7 8 9 10 11 12

Returns

Period

Returns

29.12% −23.96% 27.42% −10.09% 62.09% −19.81% −25.54% −16.53% −22.20% −6.82% 2.04% 3.60%

13 14 15 16 17 18 19 20 21 22 23 24

−18.05% 19.67% 14.34% −15.19% 6.21% 2.71% 4.10% 22.76% 39.29% 9.34% −20.34% 18.67%

The third value quoted is the un-exponentiated SD of the logs. This is ‘Latin’, and what option traders would quote. The uninitiated might note that the second and third values are quite close, and so perhaps all this is a fuss about nothing (mirroring the close VAR values in the MSCI methodology). But let’s subtly change the specification of the series above, and we shall see how wrong this complacency is. Let us suppose instead that the series in Table A2.4 is annual data rather than monthly. This is clearly a less volatile series, and obligingly we have 24 years’ worth of data. What are the three volatilities now? 1. Correct (ln) calculation 2. SD on return percentages 3. SD of log

SD=23.72% SD =22.69% SD =0.2128

Here the relative relationships have changed, with the third value now the most distant from either of the other two. This changed relationship is caused by the annualising process; it is a symptom of convexity, and is considered in the next section in more detail. Annualising When we annualise independent returns for periods less than a year, the industry universally uses the rules of the Central√Limit Theorem, which, √ as mentioned previously, boils down to multiplying monthly SD by 12 and daily SD by 262 to get annual SD.12 12 In some markets the convention is to use 250, approximately the number of working days (weekdays excluding Bank Holidays) in a year, rather than 262, approximately the number of weekdays in a year.

276

Currency Overlay

If we take our lognormal model, which is additive, then we can use the Central Limit Theorem and achieve accurate results. However, in the process of converting the log returns into percentages that we can use and which we understand, we have to exp() those log returns. In fact, ‘lognormal returns’ is a slight misnomer – so called because if you transform the return series using logs you get a normal distribution. But if you start the other way, with the normal distribution, returns are the exponential of this normal series. You could call this ‘exponentiated normal returns’. This fits the skew of returns, which is upwards. Returning to the annualising question, we√have to exp() the log series after annualising. This means that by multiplying a log (by, say, 12), we are raising the exponentiated value to a power. This is convexity in action. If all of this is confusing, a practical example might help. Suppose we have monthly data, the SD of which we wish to annualise. Table A2.5 shows the error generated by the conventional annualising process versus the log methodology. To read the table, take a value of column (a), Table A2.5 Volatility of Volatility of Annualised monthly series the logs vols √ (a) (b) = ln(1 + a) (c) = exp(b 12) − 1 1.0% 2.0% 3.0% 4.0% 5.0% 6.0% 7.0% 8.0%

0.00995 0.019803 0.029559 0.039221 0.04879 0.058269 0.067659 0.076961

3.51% 7.10% 10.78% 14.55% 18.41% 22.37% 26.41% 30.55%

Annualising % Error √ vs. ratio 12 ‘Months’ √ (d) = c/a (e) = (d/ 12) − 1 (f ) = d2 √ 12 = 3.4641 12.00 3.5070 1.24% 12.30 3.5503 2.49% 12.60 3.5940 3.75% 12.92 3.6382 5.02% 13.24 3.6827 6.31% 13.56 3.7277 7.61% 13.90 3.7731 8.92% 14.24 3.8190 10.24% 14.58

say 4%. This is the observed monthly SD, calculated by the correct log method, converted into percentages by the exp function. Column (b) shows the related monthly log SD, and column (c) the correctly calculated √ annualised SD. The remainder of the columns show the annualising ratios and the error versus 12, with column (f) showing the ratio converted into months. So for an asset with a monthly SD of 4%, this translates to an annual √ SD of 14.55%, 5% [column (e)] higher than would be calculated by a straight factor of 12 or 3.4641. This equates to √ √ 13.24 months rather than 12 months. This is a very radical finding, the importance of which cannot be underestimated for investment analysts, managers and consultants. As mentioned earlier, the one area where the problem does not arise is in option-pricing models – since no exponentiation ever appears to take place, and the traders just talk in ‘logs’, i.e. ‘Latin’. The solution Unlike many problems in investment, however, there is a simple solution. The correct methodology is available (and shown above), it is not computationally complex, and modern spreadsheets like Excel can easily compute it. The correct methodology will completely solve all the approximation problems and errors encountered in relation to SDs. Incidentally, it also resolves Siegel’s paradox (1972), which apparently gives positive combined returns for two currency market participants with opposing base currencies irrespective of the currency movement (see Box 5.3).

Appendices

277

To recap the solution, we take the observed return series, √ and instead of calculating a direct annualised SD by taking the SD % and multiplying by 12, we take √ the log(1+returns) and take the SD of this [I’ll call it SD(ln)]. We then multiply SD(ln) by 12, and calculate the √ exponential [exp(SD(ln)* 12)−1]. This is the ‘correct’ SD, expressed in % not in logs. Table A2.6 is a simplified version of Table A2.3, with the order of SD calculation clearly shown. Table A2.6 (a) Period

(b) Returns

1 2.41% 2 4.93% 3 1.40% 4 2.50% 5 1.42% 6 6.62% 7 5.53% 8 4.47% 9 2.92% 10 0.16% 3.54% 11 3.64% 12 SD monthly SD annualised SD log expressed as %

(c) Logs [ ln(1 b)] 0.023855 0.048110 0.013895 0.024721 0.014281 0.064147 0.053824 0.043688 0.028788 0.001609 0.035997 0.037083 0.033646 0.116553 12.36%

Setp 1 Calculate monthly SD of log returns Step2 annualise monthly SD (monthly 12) Step 3 Convert to percentages ( exp(SD) 1)

R REPORT APPENDIX 3 – AIMR

Copyright 1998, Association for Investment Management and Research. Reproduced and republished with permission from the Association for Investment Management and R ). All Rights Reserved. Research (AIMR

AIMR Benchmarks and Performance Attribution Subcommittee Report August 1998 The AIMR Performance Presentation Standards Implementation Committee, which sponsored the formation of this subcommittee, encourages the public to comment in writing on these standards. This report outlines guidelines concerning benchmark reporting. The subcommittee will develop a report on performance attribution at a later date. AIMR Performance Presentation Standards Benchmarks and Performance Attribution Subcommittee John C. Stannard, CFA, Chair Frank Russell Company Mary C. Cottrill, CFA California Public Employees’ Retirement System

Jeffrey P. Davis, CFA State Street Global Advisors

278

Currency Overlay

James E. Hollis, III, CFA Standish, Ayer & Wood, Inc. Robert E. Pruyne Scudder, Stevens & Clark, Inc.

James L. Kermes Glenmede Trust Company Neil E. Riddles, CFA Templeton Global Investors, Inc.

Brian D. Singer, CFA Brinson Partners, Inc.

Peter T. Willett, CFA State Street Global Advisors

Summary recommendations The committee’s report below examines some of the major factors involved in the choice of a benchmark for an investment portfolio or composite. As a result of its deliberations, the committee recommends the following changes to the AIMR standards:13 The AIMR standards should require that, where a benchmark exists, it must be included in the performance presentation. The name of the benchmark plus any other significant information (such as tax basis, etc.) must be disclosed. Where no benchmark exists, an explanation must be provided. This requirement will take effect from an effective date (e.g. 1/1/99 – to be determined by the AIMR PPS committee). Use of benchmarks for periods prior to that effective date will be recommended but not required. According to prevailing conditions (e.g. portfolio or composite strategy, index availability) a benchmark may consist of two or more indexes chain-linked together over time. If a firm introduces a new benchmark to replace one used earlier for a certain composite, it must disclose fully the reason for doing so (such as the recent availability of a better benchmark).

Introduction Benchmarks are important tools to aid in the planning, implementation and review of investment policy. They clarify communication between the investment fiduciary and the investment manager and provide a point of departure for assessing return and risk. The terms ‘benchmark’ and ‘index’ are often used interchangeably. But while indexes are most often used as benchmarks, a benchmark is essentially the starting point for evaluating success. So we might define a benchmark more generally as follows: An independent rate of return (or hurdle rate) forming an objective test of the effective implementation of an investment strategy. A benchmark may take any of the following forms: 1. A well recognised published index 2. A tailored composite of assets (or indexes) 3. A peer group (or ‘universe’) of similar funds or portfolios

13 This subcommittee has stated its recommendations in general terms rather than by specifying specific changes to specific items in the AIMR PPS Handbook. In this regard, our goal is to state recommendations clearly enough that the implementation committee has an unambiguous basis for making changes in the Handbook.

Appendices

279

What makes a good benchmark? Properly used, a benchmark should be a focal point in the relationship between the manager and the fiduciary body overseeing the prudent management of the assets. Thoughtful choice of a benchmark will make the relationship between these parties more effective and enhance the value of the investment strategy. The most effective benchmarks are: 1. 2. 3. 4. 5. 6.

Representative of the asset class or mandate Investable (e.g. a viable investment alternative) Constructed in a disciplined and objective manner Formulated from publicly available information Acceptable by the manager as the neutral position Consistent with underlying investor status (e.g. regarding tax, time horizon, etc.)

Benchmarks can be misused. Indeed choosing a bad or inappropriate benchmark can undermine the effectiveness of an investment strategy and lead to dissatisfaction between client and manager. Most problems associated with benchmarks arise from not observing the basic rules set out above (e.g. the manager doesn’t understand benchmark construction or the benchmark doesn’t match the mandate). But problems can also arise from setting multiple benchmarks which conflict with each other (e.g. outperform cash in the short term and equity in the long term). The remainder of this appendix reviews the uses of benchmarks in certain special situations and provides guidelines for investors when using and applying the AIMR Performance Presentation Standards.

Use of peer groups (‘universes’) as benchmarks The process of selecting an appropriate benchmark often involves a choice between an index (or composite index) or a peer group universe of managers. Each has advantages and disadvantages. Published indexes of ‘unmanaged’ assets are the most commonly preferred and frequently used form of benchmark, but a universe of managers may be a suitable alternative in some cases. The major advantage of using a universe as a benchmark is in situations where no widely recognised index of unmanaged assets exists to reflect the asset class or investment strategy. Examples of such situations would be real estate, private equity and venture capital. In these situations a collection of managed products rather than unmanaged assets often best represents the asset class. What is often thought of as an index for real estate, private equity or venture capital, for example, is in most cases actually a universe of managed assets or products. Universes do offer certain benefits in performance comparisons. Universes represent achieved results of manager portfolios which are effectively available as investment alternatives for investment fiduciaries (fund sponsors, etc.); they take full account of transactions and other trading costs and they reflect decisions taken by investors across the board (e.g. to underweight Japan relative to the index). But as the sole benchmark for comparing performance, universes are subject to certain drawbacks. These include: 1. They are not available real time, resulting in a time lag for comparison. 2. There is no established oversight process for determining universe participants and whether the universe accurately represents the entire asset class or style of management.

280

Currency Overlay

3. Survivor bias14 will develop over time as some managers are deleted from the universe. 4. They are not replicable or investable. 5. They do not permit the manager to move to a known neutral position. Style analysis Style analysis is an increasingly popular technique used to determine portfolio exposures to various investment ‘styles’, e.g. large-capitalisation growth. The styles themselves are generally described as passive indexes and in its broadest sense, style analysis includes the techniques used to calculate these indexes. Typically, style indexes will break a broad market index down into four (or more) mutually exclusive components usually defined as large and small (or small/medium), growth and value segments. A style analysis model will then aim to quantify the exposures of a portfolio to these four style components as expressed by the indexes (and the underlying exposures to securities within the index universes). The basic tenet of style analysis is that a passive portfolio can be constructed by combining the four indexes. Therefore, a manager can be considered to add value only when performance exceeds the passively constructed portfolio. Added value can be achieved by varying the index exposures over time or by security selection within the index universes. Application of style analysis is beneficial, but results of style analysers must be carefully interpreted. The most popular approach, developed by Sharpe,15 is a statistical technique, which assesses the styles embedded in a portfolio based on correlation with the relevant underlying index. Due to the similar performance of certain indexes, style analysers can identify index exposures where none actually exist. This is especially perplexing for global portfolios where exposures to regional indexes show up even though no assets are actually held in the region. Also, this approach is based on data measured over a fixed time horizon (say five years) and may not be sensitive to sudden changes in style. Another method assesses the style of a portfolio based on its underlying characteristics (e.g. P/E, yield, etc.). This approach is more sensitive to style changes although the model techniques must be adjusted market by market for underlying conditions and accounting measures. Style analysis results must be analysed with due regard for these factors. When using style indexes in performance presentations, investors should be careful to ensure that the style index, or blend of indexes, is representative of portfolio objectives and risk constraints. Failure to do so will lead to misleading impressions of outperformance. In this regard, the results of style models must be carefully interpreted in the light of the strengths and weaknesses of the analytical approach involved. Benchmarks for multi-currency portfolios Most investors recognise that currencies have a large impact on the returns of international and global portfolios. There is less appreciation however for the role that currencies play in the choice of an international or global benchmark. When selecting a multi-currency benchmark, the investor (implicitly or explicitly) makes both a decision on a set of underlying assets 14 Survivor bias occurs where terminated accounts drop out of the sample – hence the average or median return is biased based on the surviving (and usually better-performing) portfolios. 15 William Sharpe’s article ‘Asset allocation: management style and performance measurement’ published in the Journal of Portfolio Management (Winter 1992) presents the method now generally known as returns-based style analysis.

Appendices

281

and a decision on the desired level of embedded currency exposure. While AIMR here does not make a recommendation in favour of the unhedged or the hedged benchmark, we do regard the determination of currency exposure in the benchmark as an important fiduciary responsibility. Investors should, in addition, aim to analyse the effects of currency movements and currency decisions separately from the underlying assets. There are three basic reasons for doing so. First, currency is a large source of return and risk, both in terms of benchmark selection and active management. Second, modern performance attribution methods allow asset selection skill to be differentiated from currency selection skill. Third, currency performance and attribution results highlight the significance of the choice between an unhedged, a partially hedged, or a fully hedged benchmark. This section sets out some of the factors to consider when making the benchmark decision in the context of currency management. Note that the guidelines below are not intended to explicitly define a calculation methodology but simply highlight the factors which managers must be aware of when managing portfolios. Specific methodologies for analysing currency return must be developed in the light of underlying portfolio structure and analytical needs. Relative interest rates and the forward premium Investors often use the change in spot exchange rate (over a holding period) as a measure of the influence of currency on their portfolios (i.e. of the ‘unhedged’ currency return). However, this approach is misleading since it fails to reflect the actual returns that can be obtained by currency instruments and ignores the effect of the forward currency premium16 (referred to hereafter simply as the forward premium). In fact we get better information about the true effect of currency if we split the return derived from the change in spot rates into two separable components: 1. The forward premium which is known in advance and is driven by short-term interest rate differentials.17 2. The component of the change in spot rate not accounted for by the forward premium and which is commonly called the currency surprise.18 The reason we identify these two components separately is that they help explain an important fact. Investors cannot eliminate currency effects entirely and earn the local return.19 They can only eliminate the currency surprise component while retaining the forward premium component. Thus, the hedged return is different from the local return and is the return that the foreign investor will earn that is free of currency risk. 16 The forward premium can be positive (if interest rates in the foreign currency are lower than the home market) or negative, i.e. giving a forward discount (if rates abroad are higher). 17 The interest rate differentials are ‘short-term’ because liquidity in the forward currency market only extends to short-term contracts, i.e. up to one year. 18 The notion of ‘currency surprise’ (calculated as exchange rate movement less forward premium) is one useful basis for evaluating currency effects and currency decisions. Karnosky and Singer set out an alternative methodology in the ICFA Research Foundation publication ‘Global Asset Management and Performance Attribution’. This approach also recognises the impact of relative interest rates by adjusting nominal market returns by the risk-free rate. The Karnosky and Singer methodology is consistent with the currency surprise and forward premium framework. 19 The local return is the foreign asset return expressed in terms of the foreign currency. For example, a German equity portfolio owned by a US investor with a value of DEM100m at end-period 1 and DEM103.2m at end-period 2 has a local return of 3.2% for period 2. This return is irrespective of the spot currency movement of the Deutschmark.

282

Currency Overlay

Relative interest rates have a significant effect on currency returns, the pricing of hedging instruments such as forward foreign exchange contracts and currency analysis. While forward premiums might be thought of as a ‘cost’ of hedging, it is perhaps more accurate to consider them a ‘hedging return component’. They exist because currency-hedged returns for cash (riskless assets) are forced (by arbitrage) to be equal irrespective of the market or currency in which the investor holds the cash. So if you invest in a high interest rate, foreign cash market and attempt to lock in that higher yield by hedging the currency, the forward premium will force you to realise a loss in the currency market which is equal and opposite to the interest rate market gain. An excellent example of the currency surprise phenomenon is provided if you imagine that you are a British pound investor reviewing the performance of the pound (£) against the Deutschmark (DM) from 1972 to 1997. Commentators often claim that the pound was weak over this period since £100 converted into DM in 1972 would have been worth £253 at the end of 1997 purely through the pound’s spot depreciation. But in fact if the investor had sold DM forward for pounds, the total investment would be worth £340. This is because the cumulative forward premium (based on the relative interest rates between the UK and Germany) was greater than the pound’s actual depreciation. In fact, the currency surprise of the Deutschmark was negative by £87, and the hedged return outperformed the unhedged return.20 Accordingly, in choosing a benchmark, investors should recognise the shortcomings of using the spot-to-spot exchange rate difference. In particular, they should note the existence of the forward premium and, where appropriate, separate the spot movement into its two components – the forward premium and the currency surprise. In practical terms, this means don’t use spot-to-spot exchange rates as the basis for evaluating the success or impact of currency decisions. Unhedged benchmarks versus hedged benchmarks Unhedged benchmarks, such as the MSCI EAFE Index, are those where no adjustments are made for hedging positions. The unhedged return for each country in the benchmark consists of the combined effect of the local asset market return and the spot currency return. The benchmark therefore contains a currency component comprising both currency surprise and forward premium. Unhedged benchmarks are used when the investment mandate does not include a consistent hedged position although currency activity may be allowed as a means of adding to return and/or reducing risk. They might be used for example by investors with relatively low allocations to foreign assets and/or investors who actively desire currency exposure. 20

This example can be expressed algebraically as follows: Change in spot rate return (c) = Forward premium ( f ) + Currency surprise (s) Hedged return (h) = Unhedged return (r ) − Currency surprise (s)

In this example, c = 153%, f = 240% and s = −87% (i.e. c = f + s). Since the Deutschmark value of the converted £100 doesn’t change (i.e. local return = 0), we have: Unhedged return (r ) = Change in spot rate return (c) h = r − s = 153% − (−87%) = 240%

That is, the hedged return effectively outperforms the unhedged return.

Appendices

283

When the neutral position for an investment strategy involves the strategic hedging of some or all of the currency exposure, then a hedged benchmark (or index) is used. Such benchmarks may eliminate all currency exposure (a fully hedged benchmark) or they may eliminate a fixed proportion (between 0% and 100%) of the currency exposure. The proportion chosen is often called the benchmark hedge ratio. The return for each country in the fully hedged benchmark consists of the sum of the local asset market return and the currency premium. The currency surprise component of the spot currency return is eliminated. Once again a manager can use active currency strategies to seek additional return but the neutral position is to hold hedges equal to the benchmark hedge ratio. Hedged benchmarks might be used for example by investors with material allocations to foreign assets who wish to eliminate the volatility of foreign currencies. Benchmarking currency overlay Broadly speaking a currency overlay strategy is one in which the management of currency is carried out separately from the remainder of the portfolio – even though it may be carried out by a single manager or within a single organisation. Currency overlay is generally linked to the management of currency exposure within the portfolio – if a fund employs a currency manager but has no foreign currency assets this is not thought of as currency overlay. Currency overlay assignments can either be active or passive. Passive currency overlay versus an unhedged benchmark means neutralising the currency effect (or more specifically currency surprise) implicit in the active country position while (fully) hedged passive overlay means eliminating (all) currency surprise. Active currency overlay strategies seek to participate in upside currency gain while protecting against downside currency losses. Note in particular that, because the forward premium cannot be hedged away, hedging an asset will give you the hedged return (local plus premium) not the local return. This relationship is fundamental to the understanding of such strategies. Accordingly, the possible components of currency overlay assignment might be broken down into: (A) Hedged asset return21 (B) Currency surprise22 (C) Overlay return23

the hedged return on the assets comprising the underlying portfolio the asset-weighted currency surprise component by market the effect of the overlay strategy (i.e. active and/or passive positions) arising from the negative of the currency surprise

Such strategies can also either be fully or partially hedged (i.e. the benchmark hedge ratio can vary between 0% and 100%). These breakdowns are not significant in themselves other than that they enable us to distinguish between different components of the strategy and define benchmarks accordingly. The aim in selecting an appropriate currency overlay benchmark is that it must reflect the nature of the assignment. Since currency overlay is typically managed separately from the 21

The sum of the local market forward premium and the local asset return. Commonly referred to as the implicit currency component. 23 Commonly referred to as the explicit currency component. 22

284

Currency Overlay

Table A3.1 Description of benchmark return* Benchmark definition

Unhedged benchmark

Fully hedged benchmark

Asset + currency + overlay (A + B + C)

Hedged asset return + currency surprise (=local asset return + spot return) Currency surprise Nil

Hedged asset return (=local asset return + forward premium)

Currency + overlay (B + C) Overlay only (C)

Nil Negative currency surprise

* Partially hedged benchmark structures can be constructed as the weighted sum of unhedged and fully hedged benchmarks shown above (e.g. 50% hedged currency + overlay is 50% of currency surprise).

underlying foreign assets, performance should be correctly attributed by portfolio or manager and it is important to identify which components of the assignments are the direct responsibility of the currency overlay manager. The benchmark can be legitimately expressed either as the asset + currency + overlay return [(A) + (B) + (C)], as the currency + overlay return [(B) + (C)], or as the overlay return [(C)] only. Some examples of possible benchmark specifications are shown in Table A3.1. An investor should define a consistent currency overlay benchmark based on the strategic considerations of the fund. These considerations will include: the maturity of the fund, the size of its international asset allocation and the fund’s risk/return trade-off. The structure of the benchmark will depend on the nature of the assignment and the component of the overall strategy to be measured. Other practical considerations The following further factors must also be considered when designing currency overlay benchmarks: 1. The portion of the underlying assets that will form the basis for calculating the currency overlay return – all foreign assets, or just those for which currency hedging is practicable at reasonable cost. 2. The treatment of deviations from benchmark currency weights in the underlying portfolio. 3. Treatment of issues such as mid versus bid/offer pricing of contracts and treatment of illiquid currencies should be considered and where necessary treatment should be consistent between the portfolio and the benchmark. 4. The flow of information on the allocation of the assets being overlaid and the practical frequency for rebalancing back to benchmark – in practice, weekly or monthly rebalancing is probably more practical than daily. One-way ticket effect The one-way ticket effect describes the fact that a currency overlay manager with an unhedged benchmark can only benefit from hedging a currency when that currency is weak. Conversely, a currency overlay manager with a fully hedged benchmark can only benefit from ‘lifting’ hedges when the currency is strong. This leads to an asymmetrical, ‘one-way ticket’ influence on returns for currency portfolios. Investors should recognise that periods of continuous

Appendices

285

‘one-way’ currency movement can last for very long periods, so performance targets for shorter periods should take this into account. Partially hedged benchmarks reduce this problem, and a benchmark hedge ratio of 50% eliminates it. However, where strategic considerations make setting a partially hedged benchmark undesirable, alternatives include setting differential currency performance targets in periods of base currency weakness or strength or, probably a better course of action, allowing the currency cycle to run its course. Alternatively, an investor might set a band for currency allocation around the benchmark position, e.g. plus or minus 20% of the benchmark weight. Since, for a non-domestic currency assignment, this may give rise to a foreign currency exposure of more than 100% of the portfolio,24 this decision must be taken with due consideration for the risk parameters of the portfolio, and any legal or other limitations.

APPENDIX 4 – SAMPLE INVESTMENT GUIDELINES General remarks [CURRENCY MANAGER] will be responsible for the execution of a currency overlay programme for [INVESTOR] ‘the Client’ as follows: [CURRENCY MANAGER] will actively manage the currency overlay on the Client’s international investment portfolio comprised of [equities] [bonds] [property] [cash].

Benchmark [CURRENCY MANAGER]’s performance will be measured against an [unhedged] [50% hedged] [fully hedged] currency benchmark. See [Annex] for a description of the performance benchmark calculation. The overlay programme benchmark is determined by the Client and can be modified or adjusted by the Client upon written notice to [CURRENCY MANAGER].

Mandate size The initial mandate size is equal to the whole of the Client’s international equity portfolio as reported to [CURRENCY MANAGER] by the Custodian (see below). The Client may vary this mandate size at any time by giving written notice to [CURRENCY MANAGER].

Return and risk objectives [CURRENCY MANAGER] should achieve an outperformance on the currency overlay programme against the benchmark’s performance of [1%] p.a. net of all costs and fees over a market cycle of 3 to 5 years. A detailed description of the performance calculation is provided in the [Annex – Calculation Methodology]. [CURRENCY MANAGER] should achieve the return objective with a target tracking error of [2.5%]. The tracking error is the annualised standard deviation of the monthly outperformance of the overlay programme versus the benchmark.

24

Or less than zero for a fully hedged benchmark.

286

Currency Overlay

Eligible currencies [CURRENCY MANAGER] may only undertake spot and forward currency exchange contracts in currencies that are part of the currency overlay programme as specified below. The actively managed overlay programme will only be implemented in the following currencies: [US dollar, Japanese yen, euro, Swiss franc, Swedish krona, Australian dollar, Canadian dollar, Hong Kong dollar, Singapore dollar]. [CURRENCY MANAGER] may, at its discretion, decline to enter into hedges in any of the above currencies if either the portfolio exposure in that currency falls below a tradeable market amount, or if lack of liquidity in the market for forward currency contracts adversely affects the performance of the overlay. [CURRENCY MANAGER] may elect to hedge a less liquid currency exposure (whether or not included in the above list) by entering into forward contracts in a proxy currency (which will be in the list above). Any proxy hedging will be approved in writing by the Client prior to trading, and the list of proxies may be changed only with the written approval of the Client. An example of a proxy might be to undertake US dollar contracts to hedge a Hong Kong exposure. Proxy hedging will not be undertaken tactically – only as a strategic decision to minimise the costs of transactions. The currency composition of the active overlay programme will be designed by [CURRENCY MANAGER] and must be communicated to and agreed in advance with Client. Without the consent of the Client, [CURRENCY MANAGER] may not make any changes to the currencies that are part of the overlay programme. Investment restrictions [CURRENCY MANAGER] may only enter into forward contracts such that the hedge ratio (the foreign currency amount of the net outstanding sales of the forward currency divided by the portfolio asset value denominated in that currency) does not lie outside the range [−3%] and [103%]. Counterparties [CURRENCY MANAGER] may only enter into transactions with counterparties that have long-term senior debt or unsecured claims paying ability rated at least [Aa3 by Moody’s] and have ratings for short-term instruments of [P1] (or the equivalent ratings from other rating sources), and with the prior written approval of the Client. The creditworthiness of all counterparties must be monitored by [CURRENCY MANAGER] on an ongoing basis. [CURRENCY MANAGER] should take all reasonable steps to protect the Client’s assets in the event of adverse changes to a counterparty’s creditworthiness. It is acknowledged that it is [CURRENCY MANAGER]’s practice to remove a downgraded Bank (which falls below the permitted minimum) from the active dealing list, but to allow forward contracts to run off without premature close-out or accelerated settlement. Rebalancing The rebalancing of the overlay programme will take place on a monthly basis, on the [5th] working day of the month, except where specifically instructed by the Client, or where the Custodian has failed to provide updated portfolio data. [CURRENCY MANAGER] will normally

Appendices

287

receive portfolio information from the custodian by the [4th] working day. In case further rebalancing is to take place it is the Client’s responsibility to inform [CURRENCY MANAGER] timely about expected changes to the investment portfolio to allow [CURRENCY MANAGER] to adjust the overlay programme accordingly. Adjustments to the overlay programme will be made when the unhedged value of the underlying stock in the market relating to one currency differs from the size of the overlay in that currency by more than [3%] of the [sterling] [dollar] [euro] amount of exposure. [CURRENCY MANAGER] will set all hedges to expire on the [5th] working day of the month. Cash flows The overlay programme will periodically realise positive and negative cash flows upon settlement of the foreign exchange contracts. These settlements will be effected once a month and reduced in size if possible by using ‘settlement netting’ arrangements with the counterparties. Any cash positions may only be held in cash accounts that belong to and are controlled by the Client. [CURRENCY MANAGER] must notify and reconcile the settlement amounts with the Custodian Bank at least two business days prior to the settlement date. Any positive cash flows will be ‘swept’ from the currency overlay portfolio based upon written instructions from the Client. Cash requirements to settle foreign exchange contracts will be funded from cash sources at the discretion of the Client. Custodian The actual underlying investment portfolio will be made available electronically from the Custodian Bank. Further necessary information will be provided by the Client. It is [CURRENCY MANAGER]’s responsibility to monitor the portfolio monthly and to collect the necessary data from the Custodian and the Client. Further restrictions At no time should the net value of all outstanding contracts sold exceed 103% of the underlying value of the assets being overlaid. Forward currency contracts are limited to 13-month forward settlement dates or less. [CURRENCY MANAGER] may not deal with any affiliated firms of [CURRENCY MANAGER] as counterparty to any currency contracts. Communications [CURRENCY MANAGER] is responsible for ongoing communication with the Client. [CURRENCY MANAGER] must promptly notify the Client of any significant changes to [CURRENCY MANAGER]’s investment or portfolio strategy, organisational structure, financial condition or the personnel managing the portfolio. Reporting Monthly [CURRENCY MANAGER] will issue reports by the 12th working day of the following month as follows:

288

(a) (b) (c) (d) (e) (f) (g) (h)

Currency Overlay

Monthly overlay valuation report Monthly overlay performance report Bank credit exposure report Monthly transaction and cash flow report Monthly market summary Trailing 12-month and prospective 12-month cash flow report Cash flow at risk report Quarterly Executive Summary

The report formats will be designed by [CURRENCY MANAGER] based on the Client’s requirements. Semi-annually [CURRENCY MANAGER] will present the performance of the programme to the investment committee, and annually to the full Board of Trustees. Investment guideline compliance [CURRENCY MANAGER] will certify in writing, each quarter, that they remain in compliance with the investment guidelines, and that they have reconciled their transactions that quarter, and the outstanding contract inventory, with the Bank counterparties. If, at any time, [CURRENCY MANAGER] feels that the investment guidelines present an impediment to the investment strategy, [CURRENCY MANAGER] should discuss the matter with the Client as soon as practical. If, at any time, a specific guideline is not being adhered to, the Client should be notified as soon as possible but always within five business days after the discovery of a breach of a guideline. Addresses and contact persons Client [CURRENCY MANAGER]

Index accounting standards FAS87 11 FRS17 13, 98 IAS19 13 ACT 6, 79 active currency overlay 2, 57, 98, 100 management styles 205–37 surveys 239–40 modelling and forecasting 206–12 active management without 209–10 deal execution 212 dealing and practical execution 210 judgement and 211 timeliness of inputs 210–11 ‘adding across’ 57–60 alpha 107, 208, 211, 244 American-style options 37 analytical but realistic approach to benchmark hedge ratio 245 annualised returns 83, 85, 267–8, 275–6 annualised volatility 83 Argentina, financial regulation in 80–1 asset plus currency overlay methodology 130, 131, 136, 145–7 Association for Investment Management and Research (AIMR) 53, 125, 130, 272, 277–85 attribution analysis 58 autoregressive volatility 226 bank contract confirmation 248 bank contract settlement procedures 249 Bank for International Settlements (BIS) 66 bank foreign exchange lines 247–8 Bank of England 14 banks as customers in foreign exchange market 71, 74–5 bell-shaped curve 41 benchmarks 53, 278–9, 283–4 calculation 249–50

cash flows 141–4 constant maturity 138 currency coverage 140 current practice 144–5 definition 125–7 denominator coverage 140 design 130–44 fully hedged 282 hedge ratio 133, 139–40, 244–5, 282 investability 127–30 mechanics 126–7 misleading currency attribution 126 monthly calculation 132 for multi-currency portfolios 280–1 original contract maturity 138 peer groups (universes) as 279 performance 140–1 as portfolio 126 pricing/costs 136 purpose of 125 tracking error 148–6 unhedged 282–3 underlay 140 with/without asset returns 135 worked examples 145–8 bid implied volatilities 51 bid/offer spread 2930, 65, 68 in futures 35, 37 bilateral netting 67 BlackScholes model 40–9, 235–6 assumptions in option pricing 46–7 assumptions violations 48–9 costs in 50, 51 Greek values 44 market assumptions 40–1 option pricing 42–6 portfolio insurance story 44–6 bonds 2, 3, 182–91 correlation 182–4

290

Index

bonds (cont.) domestic 3 floating rate 3 hedged 182 index-linked 3 international diversification 187–91 stability of correlations 185–7 volatility reduction from 187 zero-coupon 190 book value 11 bottom-up management 253–4 Bretton Woods Agreement 61, 74 Buffet, Warren 210 capital account 214 Capital Asset Pricing Model (CAPM) 99, 208 capital flight 78 capital flows 62 Capital Market Line 99 carrier asset 246 ‘carry trade’ 19, 73, 195, 207, 255 central banks as customers in foreign exchange market 74–5 declared target exchange rate defence 74–5 rejection of FX intervention 75 tactical intervention 75 Central Limit Theorem 20, 40, 41, 85, 87, 196, 197, 225, 269 Chartism 74 CLS Bank 65, 67, 69 conflict of interest 257–8 constant maturity benchmarks 138 continuous adjustment 112 Continuous Linked Settlement (CLS) system 69 continuous rebalancing 112 contract rolling 129 contrarian managers 254 contribution from hedging 126, 132, 134, 143 convexity 84 corporate cash flow analysis, limitations of 162 correlations between currency returns 219–21 costs of benchmarking 136 in Black–Scholes model 50, 51 in currency swaps 33 foreign debt 23–4 forward contracts 26–31 of futures 35–7 of options 50–1 passive overlay 117–22 rebalancing 117 transactions 101–4 covered calls 48–9 cross hedging 205 Cum cash flow USD 131 cumulative cash approach 135

currency benchmark with/without asset returns 135 currency committee 70 currency coverage for benchmark 140 currency crises 19 currency exposure 57 economic impact of 21–2 embedded 163 currency futures 65 costs 35–7 margining 35 nature 34–5 sensitivity 37 currency options 38, 65 nature 37–8 pricing 38, 49–51 currency overlay managers as customers in foreign exchange market 73–4 currency overlay only methodology 130, 134, 148 currency overlay performance surveys 239–40 Currency Performance Analytics 239 currency returns 83, 158, 221 currency series 158 currency surprise 16–17, 18–19, 53–60, 126, 127, 157, 182, 201–2, 281 assets and overlay/hedging 58–60 assets in a portfolio 57 calculation 53 definition 53 geometric linking and ‘adding across’ 57–60 international assets and spot rate 58 spot rates and 54–7 currency swaps 22 costs 33 nature 31–3 sensitivity 33–4 currency unions 64 currency volatility 15–20 data mining 213 defined benefit (DB) pensions 5–6 defined contribution (DC) pensions 6 definition of currency overlay 2 delta 44, 45, 51–2, 72, 234 delta (dynamic) hedging 40, 234, 235–6 delta-hedging currency management (dynamic hedging) 212, 234, 235–6 denominator calculation for benchmark 140 discounted cash flow valuation 10 discounting 12 in MMV 139 tracking error 155 disinvesting 54 dividend taxes 79 dollar pool 15, 79 dollar pool premium 15

Index domestic bonds 9 duration 190 dynamic (delta) hedging 40, 234, 235–6 dynamic managers 212, 234–7 efficiency execution 65 semi-strong 193, 204 strong 193, 204 transactional 204 types of 193 weak form 193, 203 efficient portfolios 97 embedded currency 162 embedded currency plus currency overlay methodology 130, 133–4, 147–8 embedded dollars 164 Equitable Life 13 equities (shares) 2, 3, 157–77 base-currency-specific graphs 174–7 GBP base 176–7 German/euro base 175 Japanese base 174 US base 174 correlation evidence 158–9 historical evidence 157–8 international 169 stability of 160–2 country index equity returns 167–9 effect of hedging on portfolio risk 172–3 firm level analysis 162–6 volatility reduction 170–1 equity return series 158 equity returns 158–9 ethical managers 254 euro 21, 64 Euro 75 euro base equities 175 ‘hammock’ 180, 181 Euro-market 62–3 European Central Bank 21 European Single Currency 21 European-style options 37–8 Euro-trading 63 Eurozone 21 exchange controls 75–9 conditions for currency market liquidity 77–8 control of domestic currencies 76 forms 77 physical controls 76 price intervention 76 symptoms of controlled currency economies 78–9 Exchange Rate Mechanism (ERM) 64, 255

291

exchange rate protection 78 execution efficiency 65 expected rate of return 11–12 fair price maths 40 FAS87 11 fat tails 88–9 fixed exchange rate 62 fixed mean 224 ‘floating rate’ bonds 3 foreign assets 20 foreign debt 20–1 costs 23–4 nature 23 sensitivity 24–5 foreign exchange market banks 67 basic structure 65–70 brief history 61–5 clearing mechanism 69–70 customer types 70–5 customers 67–9 exchange controls 75–9 financial regulation 80–1 market size 66 physical and regulatory issues 75–81 taxation 79–80 turnover excluding ‘clearing’ 70 foreign exchange (FX) option writers 72 foreign exchange (FX) swaps 26, 31 forward arbitrage, boundary conditions for 263–7 forward contracts costs 26–31 nature 25–6 netting 67 sensitivity 31 forward currency prices 127 forward currency rates 51–2 forward rate bias (FRB) 198–202, 218 currency risk premium 199–200 evidence 199 monetary policy and inflation 201 nominal rate illusion 201–2 forward rate bias theory 74 fractals 226 free lunch 90–7 correlation of asset classes 94–7 currency exposure and 93–4 ‘moving parts’ 91–3 FRS17 13, 98 fully hedged benchmark 125 fundamental forecasting 74 fundamental managers modelling 212, 213–18 future value option premium 229 futures-based equitization 97

292

Index

gamma 44, 45, 234 GARCH 226, 233 Generally Accepted Accounting Principles 20 geometric linking 57–60, 129–30 German/euro base equities 175 base ‘hammock’ 180, 181 gold standard 64 growth/value 254 ‘hair trigger’ model 226 Hammerson 191 hedge funds 72–3 styles 255 hedge ratios 177–82 current debate 177–82 DEM/EUR base ‘hammock’ 180, 181 GBP base ‘hammock’ 180, 181–2 JPY base ‘hammock’ 179, 180 USD base ‘hammock’ 178–9 hedge return 126 hedge valuation USD 131 hedged bonds 182 hedged equities 157 hedged international 158 hedging instruments, range of 258 historic volatility 233 HSBC 165, 166 IAS19 13 implied volatilities 46, 51, 72, 230–1, 232 imports, rationing of 78 income tax 79 independent variables 20 ‘index-linked’ bonds 3 individual currency: equity correlations 159–60 industrial and commercial companies (ICCs) 70–1 non-profit-seeking 71 profit-seeking 70–1 inefficiency 258–9 cyclical behaviour 194 empirical evidence for medium-term trends 195–8 forward rate bias (FRB) 12, 198–202 lack of statistical arbitrage 194–5 portfolio 91 see also efficiency information ratio 207, 224, 225 Institutional Investors 3 instruments, investor 2–3 insurance companies as customers in foreign exchange market 71–2 integration 43 interest rate 51–2

differentials 136–7 swaps 22, 33 International Accounting Standards Board 20 International Bank for Reconstruction (IBRD) (World Bank) 61 International Foreign Exchange Master Agreement (IFEMA) 247 International Monetary Fund (IMF) 61 in-the-money 236 intuitive approach to benchmark hedge ratio 245 investability 127–30 investment guidelines active 245–6 alpha 246–7 passive 246 Investment Management Agreements (IMAs) 123, 248 investment managers as customers in foreign exchange market 73–4 investment pool traders in FX 72–3 invisibles 214 Japan base equities 174 yen base ‘hammock’ 179, 181 yen ‘carry’ trade 19, 73, 195, 207, 255 lagged dependent variable 218, 221 Latin Union 64 LIBID 24, 52–3 LIBOR 22, 23, 24, 52–3 life expectancy 3, 4 limit orders 122 local currency returns 56 local return 281 lognormal distribution 1 lognormal random walk 83–90, 197 measurement 83 normally distributed period returns 88–9 relevance to currency hedging 90 returns 84–6 test 89 volatility 86–8 lognormal returns 267–77 AIMR 272 annualising 275–6 industry practice 270 investment banks and option traders 273–5 literature 273–4 MSCI 270–2 return 268 solution 276–7 volatility 268–70 lognormality 49–50 Long-Term Capital Management 89

Index Macaulay duration 190 Malaysia, exchange controls and 63 management styles 253–6 active currency overlay 205–37 marked-to-market valuation (MMV) 131, 133, 138, 138 market-makers 68 Markowitz, Harry 97, 99 mean reversion 224 mean-variance optimisation 97–101 including currency 97–8 methodology 98–101 melt-downs 41 Minimum Funding Rate (MFR) 98 modern portfolio theory 99 modified duration 190 momentum managers 254 momentum model 226–7 monetary isolation 78–9 Monte Carlo models 50, 89, 196 monthly benchmark calculation 132 moving average model 223–6 MSCI 271–2 MSCI EAFE Index 141, 144, 282 multilateral netting 67 multiple regression 168 negative feedback loop with a lag 194 net capital outflows 79 netting 67 netting by novation 67 ‘no trend’ hypothesis 197 Non-Deliverable Forward (NDF) 77 normal distribution 41 normal variables 20 notional borrowing and lending 137 Occam’s razor 206 offered implied volatilities 51 oil and commodity dealers and merchants 71 one-way ticket effect 284–5 option-based managers 212, 229–34 option pricing 39–40 original contract maturity benchmarks 138 out-of-the-money 236 overlay managers active trading 232 natural selection of 256–8 passive buying 229–32 volatility forecasting 232–4 over-the-counter (OTC) 38 passive overlay 2, 98 cash flow 116–17 costs 117–22

293

cash flow management 119–21 conflict of interest and 122–3 direct 118–19 indirect 119–21 rebalancing 119 rolling 118 mechanics 105–10 benchmark/actual asset weights 108–10 currencies to be hedged 106–7 denominator of ‘contribution from hedging’ 110 frequency of asset valuation 110 frequency of cash flows 106 original maturity of forward contracts 105–6 rebalancing 111–16 buffer percentage 114–15 costs 119 delay in 115–16 frequency 112–13 valuation rates 116 Payment-versus-Payment (PVP) system 69 pension fund asset valuations 9–10 pension fund liability discount rate 11–13 valuations 10–11 pension funds 4, 9 pensions defined benefit (DB) 5–6, 7 defined contribution (DC) 6 performance measurement 250 periodic cash and contract reconciliation 249 perverse (‘upward sloping’) demand curve 72 plain vanilla options 37, 46–7 Plaza Accord 19 portfolio optimisation 98–101 portfolio volatility 92–3 present value (PV) 10 pricing/costs of benchmarking 136 probability density 42 property 191 proprietary traders 72–3 proxy hedge 109 purchasing power parity 215–17 purist approach to benchmark hedge ratio 244–5 Quantum Fund 73, 255 rate matching orders 122 rebalancing 111–16 benchmark 129, 137–8 buffer 153–4 buffer percentage 114–15 costs 119 delay in 115–16 frequency 112–13

294

Index

rebalancing (cont.) timing 154–5 valuation rates 116 regression 215, 221–3 reporting requirements 248–9 restrictions on active mandate 245–6 on alpha mandate 247 retirement age 3–4 return 266–7 returns-based style analysis 280 reverse engineering 47 rho 44 risk-adjusted excess return 225 risk-free interest rates 189–90 risk-free sovereign debt 189 risk reduction 243–4 Russell, Frank 240 scale of active mandate 245 of alpha mandate 246–7 of contracts 129 Security Market Line 99 semi-strong form efficiency 193, 204 sensitivity of options 51–2 settlement netting 67 7-standard deviation 41 Sharpe, William 225 Siegel’s paradox 87–8 simulations 227 ‘simultaneous order’ 28 smoothed value 10 Society for Worldwide Interbank Financial Telecommunication (SWIFT) 17, 18 Soros, George 72, 73, 255 stamp duty 79 standard deviation 20 ‘star’ investors 209 stock market crash 1987 45–6 straight line hedge 31 Strange, Brian 239, 240 Strange Report 239 strike rate 37 strong form efficiency 193, 204 style analysis 280 swaps 21, 22 currency 22, 31–4 foreign exchange (FX) 26, 31 interest rate 22, 33 SWIFT codes 18 tactical asset allocation 97 Taiwan dollar (TWD) proxy hedging of 109 passive overlay 101–4, 105–23

target for alpha mandate 247 on active mandate 246 taxation of foreign exchange markets 79–80 technical managers, modelling of 212, 219–28 technical trading 74 Tesco 165 theta 44 Tobin, James 80 Tobin tax 80 top-down management 253–4 tracking error 107, 130, 148–56, 250 dealing prices/benchmark prices mismatch 150–2 discounting valuations 155 hedge ratio variations 153 interest rate differential duration 149–50 in passive hedging 149–55 rebalancing trades 153–5 valuation rates for outstanding forward contracts 152–3 trade options 38 tradeable currency 108 transactional efficiency 204 transactions costs, implications of 101–4 expected portfolio added value from passive hedging 102–4 trends 197 UK cross-border portfolio investing in 14–15 pound base equities 176–7 underlay for benchmark 140 unhedge benchmark 125 unhedged currency return 135 univariate time series analysis 226 unofficial currency balances 63 ‘upward sloping’ (perverse) demand curve 72 US dollar base equities 174 base hammock 178–81 US Employee Retirement Income Security Act (1974) (ERISA) 11 US pension sector, cross-border portfolio investing in 14 valuation discounted cash flow 10 of unmatured contracts 133 value added 107, 244 variation margin 248 vega 44 very short horizon mean reversion 224 volatility 15, 20, 230, 232–4, 268–70 annualized 196 autoregressive 226

Index currency 15–20 drag 268 from equities 170–1 forecasting 232–4 of hedge equities 95 historic 232 implied 46, 51, 72, 230–1, 232 lognormal random walk and 86–8 lognormal returns and 268–70 portfolio 92–3 reduction from bonds 187

295

withholding tax 79 WM/Reuters rates 128 bank holidays 128 closing forward rates 128 closing spot rates 128 cross rates 128 intraday spot rates 128 World ex domestic index 170, 171 XY graph 153, 154 yen ‘carry’ trade 19, 73, 195, 207, 255

Watson Wyatt 240 weak form efficiency 193, 203 wide-horizon (moving average) reversion 224

zero-coupon bond 190 Index compiled by Annette Musker